Artificial intelligence (AI) has quickly become a part of daily life, influencing the information we consume and the decisions we make. And the process is just starting. This places significant responsibility on the AI builder. Designing an AI system is not merely a technical challenge but also a moral and religious one. What information is included, how it is presented and what assumptions shape its worldview all affect the end user. AI is a broad term and we speak here of any AI system that provides information or recommendations to the public, even if this is just a small subset of AI that is already integrated into systems. Judaism has long wrestled with analogous challenges, especially in the realm of publishing, where books and ideas shaped communities and beliefs. The precedents we find in halakhic literature offer guidance on the ethical responsibilities of those building AI systems today.

I. Book Publishing

When you boil the issues down to their basics, in a sense AI systems resemble book publishers. They gather, process and distribute information, often with little distinction between fact and opinion, or between traditional and secular perspectives. Of course, there are differences. Publishers determine the actual words used while AI systems have more independence in expressing ideas. However, the similarities are important. The dangers are obvious: inaccuracies can harm reputations, mislead the public and cause damage to individuals, groups or institutions. Additionally, the dissemination of a secular worldview can significantly undermine religious convictions. Judaism has a lot to say on these subjects.

But a fundamental question arises: who is the judge? Many issues cannot be conclusively proven. What counts as heretical, misleading or damaging? Who decides what is acceptable and what must be avoided? These questions, which arose in the age of the printing press, return with new urgency in the age of artificial intelligence. 

There are two ways to approach the ethical dangers of information technology: as policymakers and as citizens. Policymakers can regulate markets and restrict harmful products. Citizens, lacking that power, must find other ways to protect themselves and their communities. Halakhah addresses publishing issues from both perspectives, which can inform our discussion of AI ethics.

II. Improper Content

AI systems, even the most advanced, can generate errors. However, this is not a new challenge. Authors can include mistakes and misinformation in books, newspapers and magazines.

The Torah demands reliability. The Sages teach, chazakah she-ein chaver motzi mi-yado davar she-eino mesukan, it is assumed that a scholar does not release something that is defective and unreliable (Eruvin 32a). Your product, your words, your teaching must be accurate and responsible. This principle applies no less to an AI builder than to an author or teacher. If you release a system that frequently misinforms, you have failed the Torah standards expected of you. You might also be violating prohibitions against slander (lashon ha-ra) against individuals, groups and institutions. AI builders bear an ethical duty to ensure accuracy, reduce harm and constantly refine systems to prevent the spread of falsehoods.

But inaccuracies are not the only danger. AI can spread not only errors but also perspectives foreign and contradictory to Torah. By default, most AI systems are trained on vast libraries of secular writing, much of which reflects assumptions inconsistent with Jewish tradition. Some of these relate to unacceptable social behaviors and others relate to fundamental Torah beliefs. Presenting such perspectives as neutral fact and normative behavior and beliefs is spiritually dangerous. Books, likewise, present similar challenges.

III. Jewish Approaches to Regulating Publishing

How have Jews historically dealt with similar challenges? There are two possible perspectives: policymakers and citizens. As mentioned above, policymakers wield control and can regulate markets. But for most of Jewish history, Jews lacked such power. Indeed, Jews often utilized Christian book publishers. Instead, Jewish communities had to assert religious responsibility as citizens, finding creative ways to protect their members without market control.

Given that Jewish publishing houses have existed for centuries, it is surprising how few responsa have been published about their ethical responsibilities to the public. There is one mention of book publishers in Shulchan Arukh (Orach Chaim 307:16) declaring that the publishers of romance novels cause people to sin by thinking improper thoughts. In the 1970s, Rav Moshe Feinstein addressed the case of publishing heretical works. He famously insists that the commentary of R. Yehudah He-Chassid on the Torah is a heretical forgery. Significantly for our purposes, Rav Feinstein rules that it is forbidden for a Jewish publisher to print heresy. More strikingly, he adds that even if the overt heretical passages are removed, the publisher may not publish the rest of the work which might still contain confusing or misleading ideas. Even subtly non-traditional ideas are forbidden (Iggeros Moshe, Yoreh De’ah, no. 115).

In the AI context, this is particularly pressing. A model that offers secular or non-traditional interpretations of morality, halakhah or faith can easily mislead the unwary. The risk is not only false information but distorted frameworks of thought. AI builders must ask: what perspectives are we embedding? What worldview does the system normalize? Policymakers must consider: what perspectives can we, as a society, tolerate and what can we not? How do we enforce minimal standards to prevent dangerous views from proliferating? The first step is generating agreement that there should be minimal standards. The second step is deciding what they are. Neither step is easy. 

Even when the information comes from a reputable source, it might be improper to provide to the public. For example, the Talmud (Shabbos 30b) discusses whether certain biblical books should have been removed from circulation. There was no doubt that they were written under divine inspiration. The problem was their confusing and contradictory natures. If the objectionable passages could be explained, then there would be a basis to allow their circulation. However, responsible authorities cannot allow the circulation of a theologically confusing and misleading book, even one written under divine inspiration.

I remember when Tipper Gore led the fight against violent and profane lyrics in music. To society’s great detriment, her team’s partial win consisted only of labeling such music as explicit and nothing beyond. In my opinion, AI builders are ethically bound to ensure that AI avoids violent, profane and otherwise destructive output. And regulators are ethically bound to ensure that unethical AI systems do not enter society. However, even if this fight is won in the US, unethical AI systems will certainly be built in other countries that do not regulate their technology. Perhaps this is overly pessimistic, but it seems almost impossible to prevent those AI systems from being used in the US. In other words, no one really controls the markets. Therefore, we need to look at another model for responsible publishing.

To be continued…

This year, many people are going to try to learn lessons about how to do teshuvah from artificial intelligence (AI). I am going to go in the other direction and discuss how AI shows the wrong way to approach repentance. By studying where AI’s processes diverge from true teshuvah, we can sharpen our understanding of what real return to God requires.

I. Teshuvah and Commitment

AI learns through correction. A model is trained, tested, corrected and retrained, cycling through errors until it gradually improves. Even after training, when it interacts with the public, AI accepts corrections. Have you ever received nonsensical sources from ChatGPT? If you challenge the AI, it will acknowledge the mistake and try again. Often the next try is nonsense also. However, the AI keeps acknowledging its errors and attempting to do better. In one sense, this resembles the human tendency to stumble repeatedly before real change takes hold. 

However, Rambam defines true teshuvah differently. In Mishneh Torah (Hilkhos Teshuvah 2:2), he writes that teshuvah is reached when a person commits to never repeating the sin. He adds that the Almighty Himself testifies the sinner will not return to that transgression. This seems like quite a high burden. Ostensibly, it means that you have not accomplished teshuvah unless you never again become weak and fall prey to this sin. In other words, your teshuvah is always conditional until the day that you die.

The implausibility of this reading leads commentaries to explain Rambam differently. Rav Avraham de Boton (16th cen., Greece) writes that Rambam means that you call down God to testify to your sincerity (Lechem Mishneh, ad loc.). Rav Yitzchak Blaser (19th cen., Russia) explains that God testifies that if all things remain equal, based on current circumstances this person will not return to this sin (Kokhevei Or, no. 7). Regardless, Rambam requires a sincere commitment to refrain from this sin going forward.

This is not iterative, like AI’s process, but decisive. AI’s endless cycles actually highlight how people often fall short of full teshuvah, circling through half-measures and repeated failures. People might intend to refrain from sin but they usually lack commitment to meaningfully change their lives. However, true teshuvah demands a higher standard: an unwavering break with the past, a level of transformation so deep it achieves divine affirmation. In this sense, AI represents the common half-measure of teshuvah rather than the necessary full teshuvah.

II. Humility and Focus

In other respects, however, AI can teach us much about teshuvah. Despite its impressive capabilities, AI never claims perfection. It always functions within boundaries of probability, with an inherent awareness of error margins. A system that pretends to be flawless is misleading and dangerous. Teshuvah begins with the same humility, the same awareness of self-limitations. A person who insists on his perfection, who refuses to acknowledge failure, cannot change. Recognizing imperfection is not weakness but strength. It opens the door to change, to growth, to excellence. The important lesson for teshuvah from AI is that accepting your imperfections does not demonstrate weakness. It allows for meaningful improvement.

King David says in Tehillim: “One thing I asked of the Lord, that which I will seek: That I may dwell in the house of the Lord all the days of my life…” (Ps. 27:4). There is a special significance not just in what David requests but in that it is only one thing. Normally in life we strive for the next world but pray for this world. The Gemara (Shabbos 10a) tells how Rava saw Rav Hamnuna, who was spending a long time in his prayers. Rava said about him: “They abandon eternal life (chayei olam, i.e. Torah study) and engage in temporary life (chayei sha’ah, prayer).” When we learn and do mitzvos, we acquire eternal merit, we focus on the long term spiritual life. When we pray, we ask for things we need in this world, we focus on the short term. Rava criticized Rav Hamnuna for praying too long when he could have been learning.

The Gemara explains that Rav Hamnuna set different times for prayer and for learning. However, the continuation of the Gemara makes it seem that Rava’s argument is conclusive. R. Yirmiyah and R. Zeira were learning Torah. Prayer time arrived and R. Yirmiyah rushed to begin his prayers. R. Zeira criticized him, quoting the verse, “One who turns his ear from hearing Torah, even his prayer is an abomination” (Prov. 28:9). According to Rava and R. Zeira, we have to pray but we should minimize the time we spend in prayer so we can learn more Torah. While we have to live in this world, our focus should be on the next world.

III. Teshuvah and Focus

Rav Avraham Shmuel Binyamin Sofer (19th cen., Hungary) explains that David’s focus was directed to the next world. The one thing he requested was to live a life of spirituality, focused solely on the next world, not on prayer but on Torah and divine wisdom. No person can reach that on his own but if we put in our effort, God will help us reach it (Kesav Sofer Al Ha-Torah, Inyanei Teshuvah, p. 729-730 in the 1995 edition). The proper spiritual attitude is one of focus on the spiritual life. In this, we can learn something from AI.

AI is relentlessly focused on its assigned task. It does not tire, become hungry or get distracted by bodily desires or idle curiosity. Once given a goal, it directs all its energy toward achieving it. This provides a sharp contrast to human life, where distractions abound and our spiritual goals are often sidelined by comfort or habit. Teshuvah requires recovering that clarity of focus. Torah and mitzvos define the ultimate purpose of life, but staying directed toward them takes conscious effort. AI’s single-minded drive reminds us of the concentration we should bring to avodas Hashem.

None of this implies that AI is capable of spirituality. Machines cannot repent. However, their structures and patterns echo truths about human growth. Iteration, which falls short of teshuvah, shows how people often drift into cycles of repeated sin. Humility teaches openness to correction. Focus teaches determination. Without the willingness to change, people not only remain flawed but descend further. A car that is not regularly serviced breaks down. Similarly, a person who does not regularly reevaluate and adjust his conduct, devolves into antisocial behavior. AI, a creation of human ingenuity, unintentionally reflects some of the same principles that sustain spiritual life. It shows us where we can go wrong, and also where we can go right: committed, humble and focused correction leads to lasting transformation.

I. AI and Unemployment

Predictions vary widely about the impact of artificial intelligence (AI) on employment. All agree that AI will replace some jobs, particularly those open to the automation of processes. The critical question is whether AI will become a tool or an independent actor. If it remains a tool, AI will give more power to employees and may even create more jobs than it displaces. If it becomes independent, it might lead to widespread unemployment as AI provides more efficient, more accurate and more economical work. My inclination is that the former will happen but it is worthwhile to explore the second possibility and its religious significance.

You might remember when Andrew Yang ran for president in 2020 that part of his platform was a Universal Basic Income (UBI). Among other things, UBI would guarantee workers displaced by AI a minimal income so they can cover basic expenses. People will still be able to rise above this minimal level through work and initiative, but those unable or unwilling to do so will not starve. Yang did not win his party’s nomination and UBI is still an eccentric political proposal. If the second scenario above occurs, this proposal might become a reality. Thinking about UBI helps illuminate an interesting teaching of the Rambam.

II. The Natural Order

In multiple places in the Talmud, we see a debate about the Messianic Era between the sage Shmuel and the other rabbis. According to R. Yochanan, the prophets, in their discussions of the miraculous good awaiting us in the future, were discussing the Messianic Era. However, the World-to-Come, the afterlife, remains a mystery to us. Shmuel believes that the Messianic Era will not see the fulfillment of the prophecies. Rather, the only difference between our current existence and the Messianic Era is that in the future we will not be subject to foreign governments. The world will still continue to follow the natural order.

The challenge is that Rambam seems to follow both views. In one place in Mishneh Torah, Rambam writes that the prophets did not discuss the World-to-Come (Hilkhos Teshuvah 8:7). Meaning, he follows R. Yochanan. However, later he writes (op cit., 9:2, Touger translation):

”For these reasons, all Israel, [in particular,] their prophets and their Sages, have yearned for the Messianic age so they can rest from the [oppression of] the gentile kingdoms who do not allow them to occupy themselves with Torah and mitzvot properly. They will find rest and increase their knowledge in order to merit the world to come… [Nevertheless,] the ultimate of all reward and the final good which will have no end or decrease is the life of the world to come. In contrast, the Messianic age will be [life within the context of] this world, with the world following its natural pattern except that sovereignty will return to Israel. The Sages of the previous generations have already declared: ‘There is no difference between the present age and the Messianic Era except [the emancipation] from our subjugation to the [gentile] kingdoms.’”

In this passage, Rambam follows Shmuel, who says that the natural order will continue in the Messianic Era and the only change will be freedom the domination of other nations. He even repeats this point at the very end of Mishneh Torah (Hilkhos Melakhim 12:2). Yet earlier Rambam rules like R. Yochanan. How do we make sense of this?

Rav Avraham de Boton (16th cen., Greece) suggests that Shmuel and R. Yochanan are not that far apart. According to R. Yochanan, as Rambam reads him, the prophecies refer to the Messianic Era but are meant allegorically. Both R. Yochanan and Shmuel agree that the natural order will continue in the Messianic Era. However, R. Yochanan believes that there will also be world peace and universal acceptance of God. In contrast, according to this approach, Shmuel believes that the only change will be governmental. Rambam follows R. Yochanan but still is justified in saying that the natural order will continue and that the main difference will be governmental (Lechem Mishneh, Hilkhos Teshuvah 8:7).

Rav Nachum Rabinovitch (21st cen., Israel) suggests an approach that is both similar and opposite. According to Rav Rabinovitch, Shmuel and R. Yochanan both agree that the prophets were discussing the Messianic Era. However, R. Yochanan takes the prophecies literally while Shmuel understands them allegorically, within the natural order. Therefore, Rambam clearly follows Shmuel (Yad Peshutah, Hilkhos Teshuvah, introduction to ch. 9).

III. AI and Torah Study

These two approaches represent opposite poles of approaches to reconcile Rambam’s rulings. Other attempts have been made, somewhere in between. Regardless, it seems clear that according to Rambam, the Messianic Era will not be particularly miraculous. Rather, it will be a time when people expend much effort in studying Torah and apprehending God’s ways.

The question remains how this can occur within the natural order. Where will people suddenly find the time to study Torah during their busy lives? Perhaps AI provides the answer. If the second scenario described at the beginning of this essay takes place and AI replaces most jobs, leading to a UBI, nearly the entire country— and perhaps world—will be free to spend their days studying Torah and contemplating divine truths. Ironically, the very technology that many fear will take away purpose could become the enabler of the Messianic dream. Thanks to AI, the world will learn in kollel.

I. The Cost of Secular Education

The high cost of yeshiva tuition remains a significant challenge to many in the community. Artificial Intelligence (AI) promises a potential solution: it can reduce costs while simultaneously improving the education. Personalized instruction without the staff overhead make AI a potential game changer for yeshiva secular studies, provided it is designed properly.

For example, AI-based tools can provide personalized learning programs, giving each student their own content and learning pace. Where a human teacher may be responsible for dozens of students, AI can deliver individualized instruction simultaneously to all, with a teacher overseeing many classes as a supervisor and supplemental resource rather than as the primary educator. This technology would dramatically lower the need for staff, allowing schools to reduce payroll, the single largest driver of tuition costs.

It remains to be seen whether AI can serve effectively as a teacher. We must also ask whether halakhically we are allowed to use it in this function.

II. Teachers and Yeshiva Education

The Talmud (Avodah Zarah 15b) rules that one may not send a Jewish child to a gentile teacher. The reason for this prohibition is debated by commentators. Rambam (Mishneh Torah, Hilkhos Issurei Bi’ah 22:5) explains that the concern is primarily that gentile teachers in the ancient world were free to physically abuse Jewish students. If that danger is absent, Rambam implies, the prohibition does not apply. Rashi (ad loc., s.v. ve-ein mosrin lahem), however, presents a dual concern: not only the possibility of physical harm but also the spiritual danger that the teacher might influence the child to stray from traditional Jewish belief and practice. According to Rashi, the concern for indoctrination remains even if physical abuse is no longer an issue.

Rav Moshe Isserles (Rema; Yoreh De’ah 153:1) follows Rashi, which the Vilna Gaon (glosses, ad loc.) explains is because the concern about improper influence persists. Rav Shmuel of Furth, in his Beis Shmuel commentary on Shulchan Arukh (Even Ha-Ezer 22:7), argues that Rav Yosef Karo likewise follows Rashi. Thus, we find a critical disagreement: Rambam permits gentile teachers under conditions of safety, while Rashi maintains the prohibition due to the risk of ideological influence. Modern authorities follow Rashi’s strict view.

This seems to mean that yeshiva day schools should be forbidden to employ gentile teachers for secular subjects. And yet this is common practice. The explanation, I believe, lies in the difference between the ancient model and the modern school system. In Talmudic times, a child was sent to a teacher who functioned independently, teaching according to his own ideas. Such an arrangement raised both dangers outlined by Rambam and Rashi. Today, by contrast, secular teachers are employed within institutional frameworks. The yeshiva sets the curriculum, supervises instruction and enforces proper behavior. Teachers may not introduce ideological material outside of the approved syllabus.

I have found little discussion of this topic among the major halakhic authorities. The earliest I had found is Rav Akiva Schlesinger, in his 19th century polemic, Lev Ha-Ivri (vol. 1, p. 29n1). In his opposition to non-Orthodox Jews, Rav Schlesinger argues that those who are legally obligated to provide secular studies should hire gentile teachers rather than non-Orthodox Jews because the latter will be more likely to proselytize. In a footnote, he points out the Talmudic prohibition and argues that it only applies when a child goes to the teacher’s home or school to study. But when the teacher comes to the Jew’s home or to a Jewish school, it does not apply.

Rav Avraham David Horowitz (20th cen., Israel) reached an even more lenient conclusion. He argues that the prohibition only applies when there is a concern that a teacher will taint his class with his personal beliefs. However, if a secular school maintains standards and professionalism, a Jewish student may attend without concern (Kinyan Torah Ba-Halakhah, vol. 1, no. 55, par. 7). Of course, nowadays secular universities allow professors free reign to dilute their courses with their secular ideologies so no such permission remains.

Jewish day schools do not allow professors to inject their personal beliefs into the classroom. The Jewish studies faculty serves as role models and life guides while the secular studies faculty is there to teach the syllabus. This practice would halakhically justify the common practice of Jewish schools hiring gentile teachers for secular studies. However, those schools that, in the name of academic freedom, do not restrain their teachers would presumably fall under this prohibition.

III. AI as Secular Teachers

When applied to AI, this distinction becomes critical. According to Rambam’s view, an AI program cannot physically harm a child. The central concern—violence—does not exist. Therefore, Rambam’s framework would support permitting AI to serve as a secular studies instructor. According to Rashi’s approach, the concern of ideological influence remains. While AI has no beliefs of its own, its training data and algorithms may embed non-traditional or secular worldviews into its functioning. These could subtly shape a student’s thinking, precisely the concern Rashi articulated. On this basis, an unsupervised AI would be prohibited. Since the Rema codifies Rashi’s emphasis on ideological influence, the halakhic presumption is that AI instruction would be forbidden unless adequate safeguards are in place.

However, if supervision addresses the concerns with gentile teachers, it should likewise address the risks of AI. An AI system can be restricted through technical and educational “guardrails.” Programmers can filter its content, block religious or ideological commentary, and confine its role to secular subjects such as mathematics, science or language. It can even push it in the direction of traditional Jewish beliefs, forcing the AI to adopt an Orthodox viewpoint. In addition, rabbinic and educational supervisors can test the system’s outputs to ensure conformity with traditional Jewish standards. In fact, AI may present fewer risks than human teachers. While human instructors inevitably bring personal worldviews into the classroom, AI can be explicitly designed and constrained. It lacks independent agency. Thus, when properly programmed and supervised, AI arguably offers a safer and more controllable means of delivering secular education.

For yeshivos under pressure to contain tuition costs, this understanding opens a significant opportunity. If AI can be implemented with the necessary safeguards, not eliminating secular teachers but reducing the head count significantly, schools may deliver secular education more effectively at far lower cost. This could alleviate the crushing burden on families while preserving the high standards of Torah education. By investing in oversight and careful design, the community can reduce the financial burden while enhancing commitment to tradition.

OpenAI recently released ChatGPT-5, which faced a number of public challenges. One of its features is that users can now choose from five preset Artificial Intelligence (AI) “personalities.” The list of personalities includes cheerful and adaptive; thoughtful and supportive; exploratory and enthusiastic; efficient and blunt; and, most provocatively, critical and sarcastic. In other words, you can now speak to an AI designed to respond with skepticism, sarcasm and biting commentary.

This raises an ethical and halakhic question. If AI is only code generating text, do its character traits matter? Or can it subtly influence us, shaping how we speak and even how we think?

I. Growth and Influence

Human beings are social creatures who continually evolve, with their thoughts and actions shaped in part by the company they keep and the models they observe. In order to grow in the right direction, we must surround ourselves with positive role models. Rambam vividly explains the proper attitude to this psychological reality (Mishneh Torah, Hilkhos De’os 6:1):

“It is natural for one’s character traits (de’os) and actions (ma’asim) to be influenced by friends and companions and for one to follow the local norms of behavior. Therefore a person must associate with the righteous and constantly sit with the wise in order to learn from their deeds. And he should distance himself from the wicked who walk in darkness so that he will not learn from their deeds.”

This statement is not just general guidance but has specific practical implications. For example, Rav Shlomo Aviner (cont., Israel) writes in a responsum that Israelis should not spend extended time in India after their army service because, among other reasons, it involves being in close quarters with non-religious and non-Jewish people whose behaviors can have a profound negative influence (Responsa She’eilas Shlomo 4:246). Similarly, Rav Shlomo Min Hahar (20th cen., Israel), in his 1971 halakhic guide for IDF soldiers (Dinei Tzava U-Milchamah, par. 5) cautions that they should not spend downtime with non-religious soldiers because of the potential impact on their spiritual character.

However, a subtle shift in Rambam’s language might have significant relevance for our question. He begins with both de’os and ma’asim—attitudes, dispositions, character traits and deeds, behavior, actions. He concludes that we must keep company with the righteous and avoid the wicked because of their deeds. Why the move from traits to behaviors?

II. The Power of Traits

Given the context of Hilkhos De’os, which discusses the proper character traits, it is difficult to sustain the argument that halakhah primarily governs what we do, not what we silently feel. It seems more likely that Rambam frames the issue in behavioral terms because character traits are abstract until they appear in conduct. A selfish attitude becomes contagious when it is displayed in action. It is true that we must be worried about adopting the unhealthy character traits of those who surround us. However, Rambam highlights actions because generally they are the vehicle through which traits spread.

Alternatively, perhaps Rambam feels that from a practical standpoint, we need only be concerned with being influenced by other people’s concrete bad deeds and not their intangible attitudes. We cannot be influenced materially by the thoughts inside other people’s hearts. Our focus must be on maintaining positive deeds, which will in turn sustain positive attitudes. If we take care to avoid unhealthy actions, we will avoid being influenced by other people’s negative character traits.

This last approach raises the important question: if Rambam focuses on actions, does that mean that we need not worry about being influenced by other people’s traits if they are not translated into action?

The Gemara (Shabbos 33a) teaches that one who merely hears nibul peh, vulgar or profane speech, is punished. He himself did not speak improperly. However, the words alone, encountered passively, carry moral harm. This text implies that exposure to speech itself, without physical action, can corrode character. Even language divorced from deed conveys a personality trait and influences the listener. If vulgar speech damages, then cynicism, sarcasm or arrogance transmitted in words can also damage.

III. AI and Character

With this in mind, we can return to our discussion of AI personalities. A human being has both traits and behaviors with influence often flowing through deeds that then shape traits. An AI chatbot has no deeds; it does not act in the world. Its entire presence is in speech.

This means that the danger lies not in AI’s actions, because it has none, but in its simulated traits. An AI cynic models cynicism. An AI with a sharp tongue normalizes sarcasm. A dismissive or arrogant personality slowly habituates its users to those same traits.

Must we worry about AI’s corrosive influence on character traits? This seems to depend on how we understand Rambam’s shift in language from de’os and ma’asim, traits and actions, to just ma’asim, actions. If we believe that Rambam is concerned for both but focuses on actions because generally that is how attitudes are transmitted, then we must avoid a chatbot’s corrosive speech. On the other hand, if Rambam only warns us about learning from others’ negative deeds because attitudes are not transmittable, then we need not be concerned about an AI chatbot’s personality.

A third possibility is that speech constitutes action. Traditionally, there is a debate whether speech rises to the level of action. For example, the Torah forbids muzzling a working animal to prevent it from eating (Deut. 25:4). If you effectively muzzle an animal by verbally berating it, R. Yochanan says you violated the prohibition through speech while Resh Lakish says that speech does not constitute action (Bava Metzi’a 90b). Tosafos (ad loc., s.v. R. Yochanan) quote a number of relevant texts regarding whether speech constitutes action and conclude that speech constitutes action if it yields tangible results, like an animal refraining from eating. In our case, it seems farfetched to say that an AI’s words lead to action and therefore that the speech constitutes action.

IV. Practical Implications

The upshot is that creators of AI tools, particularly but not exclusively for Jewish use, must design them to speak with derekh eretz and kavod, manners and respect. AI must model responsible behavior so it becomes a positive, and certainly not a negative, influence.

When this is not the case, according to one reading of Rambam, we are instructed to distance ourselves from bad influence. Choosing a sarcastic or cynical AI personality is no different from choosing friends who mock and belittle. Even if the AI is only computer code, the influence of its words is real. Interacting daily with an AI cynic or a sarcastic assistant risks habituating yourself to bad traits. This is not harmless fun; it is a steady shaping of character.

Rambam writes above that “a person must associate with the righteous and constantly sit with the wise.” Today, that advice applies not only to our neighbors and friends but also to the personalities we allow into our lives electronically. AI can be a tool for Torah and personal growth but only if it speaks with dignity and refinement. Otherwise, it becomes a daily tutor in bad midos.

I. AI and Transformation

Artificial intelligence (AI) is changing the world and already beginning to transform the workplace, a process that is widely expected to be dramatic. Will Torah study face the same fate? Some anticipate that AI, once it overcomes its current accuracy problems, will transform how we learn: whether replacing the rebbe (teacher) or chavrusa (study mate), reimagining what a text is and looks like, or offering individualized courses of study based on ability and interests. Perhaps more deeply, some think that not just the form of Torah study but the very goal will change also. Do we need to study texts carefully when we can easily obtain all our answers from AI? Maybe the curriculum should change to more about personal growth and less about textual mastery.

Yet history teaches us that the core of Talmud Torah does not bend so easily to technological disruption. Torah is not simply information to be processed. Advanced Torah learning is itself a spiritual discipline, a form of avodas Hashem. The connection with a rebbe and the yegi’ah and amalah, the effort and toil, are part and parcel of personal growth as a Torah Jew. These are not incidental features of Torah study; they are the essence of it. Without the sweat, the back-and-forth argument, the human relationship and the process of internalizing Torah’s values, the words remain external, unintegrated into one’s mind and heart. No machine can substitute with the process of becoming a part of the chain of Torah transmission throughout the generations.

To understand AI’s place in the future of Torah study, it helps to recall a much earlier and far greater technological disruption: the writing down of the Talmud.

II. The First Great Disruption

For centuries, Torah She-Be’al Peh, the Oral Torah, was transmitted exclusively through memory and speech. Students learned by listening, repeating and reviewing constantly. A Torah scholar was, above all, a living library. Mastery meant knowing vast bodies of material by heart and being able to recall them instantly in debate or judgment. When the Talmud was redacted and committed to writing, this world changed. The act was unprecedented and far more revolutionary than AI is today because it fundamentally altered how Torah was accessed and preserved. What had been stored in the minds of sages could now be stored on parchment.

This shift also affected the halakhic requirements for Torah knowledge. The Sages speak intimidatingly about the prohibition against forgetting the Torah. The Mishnah says: “whoever forgets even one thing of his learning, Scripture considers him as if he is liable for his life” (Avos 3:8). Rav Shneur Zalman of Liadi (19th cen., Russia) discusses this prohibition at length, expanding on the details (Shulchan Arukh Ha-Rav, Hilkhos Talmud Torah 2:4-8). However, his contemporary, Rav Chaim Volozhiner, argues that this is no longer applicable. This prohibition was for a time when the continuity and very existence of the Oral Torah required people memorizing and retaining their learning. A failure to remember posed a danger to the tradition itself. Nowadays, with the material preserved in writing, forgetting a detail no longer carries the same consequence, since it can be recovered from the text (Keser Rosh, no. 67).

The emphasis of learning also shifted, to a degree, from memorization to textual and abstract analysis. Scholars still value broad knowledge, but the primary skill has become understanding and interpreting texts, rather than holding every line in memory. A good memory is still held in esteem but more as a curiosity than a requirement. The change lessening the value of memorization did not diminish Torah study; it enriched it.

III. AI as the Next Step

Seen against this backdrop, AI is not an existential threat but a continuation of the same trajectory. Just as the written Talmud shifted the center of gravity from memory to analysis, and just as digital search tools made it easier to locate sources, AI will enable certain tasks that previously were unavailable to the average student. It will not, and cannot, change the process of learning itself.

The ways in which AI can serve as a tool in Torah study are still being discovered as the AI revolution begins. Here are at least three important ways AI can be used in learning:

1) Source Discovery and Mapping – Imagine an app with which you take a picture of a text and AI instantly locates relevant sources across texts, commentaries and codes, tracing where they are cited in Talmud and later literature, and identifying modern texts, articles and responsa on the same topic. This allows a learner to see not only the origin of an idea but also its development and application over centuries.

2) Topic Summarization – AI can produce concise overviews of any sugya or halakhic topic, linking directly to the primary sources for deeper study. A student attempting to review and digest a large topic can quickly obtain an organized view of the relevant material he has learned.

3) Historical Context – AI can provide background on the era, geography and culture referenced in a source, helping clarify difficult passages that assume familiarity with ancient realities. This will particularly enhance the study of Aggadata (non-legal narratives) but also help students understand Talmudic, medieval and early modern texts about economic and social activity.

Tools like these offer exciting opportunities to enhance the learning of students and scholars who have already mastered the basic textual skills necessary for learning. They will not replace classical Torah learning but supplement it.

IV. Breadth of Knowledge in the AI Era

With AI able to find and retrieve almost any text in seconds, one might conclude that broad knowledge will become obsolete. But just as in the post-redaction period, breadth will remain indispensable. The student needs to learn and master the entire Torah in order to internalize it, to allow it shape his thinking and worldview. Doing so allows you to recognize patterns, draw connections and respond instinctively through a Torah lens.

AI retrieval mirrors the historical shift: it reduces the practical necessity of memorizing every detail, but it raises the bar for broad knowledge, deep reasoning and analysis. A related discussion appears in the Talmud about the relative value of perfect recall versus penetrating analysis. The Gemara (Horayos 14a) describes a debate whether the great yeshiva in Pumbedisa should choose Rabbah or Rav Yosef as its rosh yeshiva. Rav Yosef had a phenomenal memory and knew all Tannaitic literature by heart. He was a “Sinai.” Rabbah did not have a similarly encyclopedic knowledge but was a brilliant analyst, an “oker harim” (uprooter of mountains). Who is a more appropriate choice for rosh yeshiva? Rav Yosef, the Sinai, was selected but he declined and Rabbah took the position. Twenty two years later, after Rabbah’s death, Rav Yosef assumed the position.

Rav Shlomo Kluger explains that this conclusion reflects the time before the Oral Torah was written, when a Sinai’s memory was indispensable. Once the texts were committed to writing, the advantage shifted toward the oker harim (marginal note to Pri Megadim, Orach Chaim, Eishel Avraham 136). In our day, with Google, Bar Ilan and digitized Jewish libraries, this argument becomes even stronger: analytical skill may be more critical than encyclopedic memory. Still, as Rashi notes (Horayos 14a s.v. u-mar), an oker harim must also possess some Sinai, i.e. familiarity with all the sources, even if not instant recall. The novice analyst is no substitute for the seasoned scholar. A sound analytical thinker must know the entire Torah to internalize the concepts and attitudes.

V. The Unchanging Core

In the end, AI will be an additional tool, not a new method. The essential elements of Torah study will remain. Technology can help us find the material and arrange it neatly on our desks, but only human effort can turn it into wisdom and holiness. Evidence of this minimal impact can be found in the growth in popularity of Daf Yomi textual study and even memory retention skills, which took place at the same time that the Internet grew and a variety of websites and apps for Torah study became available. There is a draw to classical Torah learning that supersedes the availability of technology, even as technological aids are used to assist with and supplement the classical study of Torah texts with a rebbe or chavrusa.

While the printing press, the Internet and now AI have had great impacts on Jewish society and Torah learning, the greatest transformation in Torah study already happened some 1,500 years ago when the Talmud was written down. Every other development since has been a smaller step in the same direction. AI is no different. It will accelerate research, open up new connections and make some kinds of work easier. But the heart of Talmud Torah remains exactly where it has always been: in the discipline and hard work of those who choose to engage in this sacred act of worship. AI will enhance the hard-earned learning obtained through yegi’ah and amalah, supplementing our classical learning with new tools.

Artificial Intelligence (AI) is changing how we approach complex problems across a wide range of fields. From diagnosing rare diseases to reconstructing lost languages, AI has proven itself capable of analyzing vast amounts of data, detecting patterns and offering insights that human researchers might miss. Perhaps there is room to ask whether AI could help solve a very intricate and complex religious question: can it determine the exact location of the mizbe’ach, the altar in the Beis Ha-Mikdash, the holy Temple in Jerusalem?

At first glance, the question seems technical. Feed AI everything we know from the Bible, Sages, commentators, ancient and medieval attestations, and archaeological surveys, and let it find patterns and connect information in order to identify the precise location. Many researchers have attempted to do this over the past century, yielding conflicting results. Perhaps AI can reach a conclusive determination that satisfies all parties. However, halakhah is not always governed by evidence alone. Sometimes it demands a specific type of certainty, one rooted not in analysis, but in tradition or prophecy. The placement of the mizbe’ach may be such a case.

I. A Fixed and Holy Location

The mizbe’ach is not just another vessel in the Temple. Its location is permanent, not subject to change. The Rambam writes: “the altar’s place is very precise which may never be changed” (Mishneh Torah, Hilkhos Beis Ha-Bechirah 2:1). This spot is the same location where Adam brought his sacrifice, where Noach offered sacrifices after the Flood, where Avraham built the altar at the Akeidah and where David and Shlomo ultimately built the Temple (ibid., 1-2).

The Talmud (Zevachim 62a) emphasizes the need for precision in placing the altar, ruling that it must be built on its exact location. This is not merely a preference for historical continuity, but a halakhic requirement embedded in the sanctity of the site. Building the mizbe’ach even slightly off-site would invalidate the sacrifices offered on it.

II. Prophetic Identification

How was the correct location identified? The Bible describes how the prophet Gad instructed King David to purchase the threshing floor of Aravna the Jebusite and build an altar there (1 Chron. 21:18–30). This episode is not simply a historical footnote, but rather serves as the halakhic basis for the location of the Temple. Rambam’s citation of this story emphasizes that the placement of the mizbe’ach was confirmed through prophecy.

In 1862, Rav Tzvi Hirsch Kalischer published his Derishas Tziyon, in which he advocated for the immediate resumption of sacrificial worship. In the third essay (part 1, par. 2, p. 91 in the 2002 edition), Rav Kalischer says that Rav Akiva Eiger asked him how we can know where the altar should be if in the past they needed prophets for that. Rav Kalischer replied that we can measure from the Western Wall and recreate the Temple based on the Mishnah tractate Midos.

Rav Kalischer’s colleague, Rav Eliyahu Guttmacher, challenged him that we need a prophet to identify the place for the altar (p. 134). Rav Kalischer replied that the prophecy was needed not to identify the place of the altar but because they wished to expand the altar and needed the prophet’s instruction how to do that. However, reconstructing the altar itself requires only knowledge of its past place. Rav Yechiel Michel Tukaczinsky answers more simply that King David did not know where the altar should be built and require a prophet to show him. If we can determine the proper place without a prophet, we are allowed to build the altar on our own (Ir Ha-Kodesh Ve-Ha-Mikdash, vol. 5, ch. 6, par. 7, p. 67).

Rav Yaakov Ettlinger argues with this view (Binyan Tziyon, no. 1). He contends that even if we know the exact site of the altar, we need prophetic permission to offer sacrifices there. Many others follow suit, making this the majority, mainstream view (see Rav J. David Bleich, Contemporary Halakhic Problems, vol. 1, pp. 259-262). According to this approach, we cannot rebuild the altar or the Temple, nor offer sacrifices, without the instruction of a prophet.

III. Prophecy or Evidence?

Particularly since the Six Day War’s reconquest of the Temple Mount, many studies have been published attempting to determine the location of the Temple, with conflicting conclusions. AI could potentially play a significant role in resolving any doubts that remain. Its capacity to synthesize texts, analyze spatial data and correlate seemingly unrelated pieces of evidence makes it uniquely suited to the task of finding the right spot. AI could process all references to the Temple Mount in rabbinic literature, compare descriptions across centuries, overlay historical maps with modern satellite images and match archaeological findings with textual data. Such a system might determine the precise location of the altar relative to other fixed points. It might even be able to reconstruct a highly plausible model of the Second Temple and superimpose it on the current Temple Mount. In a purely analytical sense, AI could offer the most precise determination ever produced. But even if its conclusions are correct, are they halakhically actionable?

According to Rav Kalischer, AI can help us sort through the information and make a detailed plan for the altar and the Temple. All we need is certainty, whether it comes from prophecy or analysis. However, according to Rav Ettlinger, prophecy is needed to rebuild the altar. Even the most advanced AI cannot resolve the halakhic uncertainty and offer us permission. Halakhah might treat all AI conclusions as impressive but ultimately insufficient. According to this approach, neither history nor science, archeology nor architecture, can allow us to build the altar.

IV. Recreating Halakhic Items

This tension between evidence and authority might be broader than reconstructing the Temple architecture. For example, a comparable debate exists regarding the identification of tekheiles, the colorful dye used in tzitzis. After centuries without it, some contemporary researchers have argued that the chilazon, the sea creature that produces the dye, has been rediscovered in the murex trunculus.

Many halakhic authorities, including Rav Hershel Schachter and Rav Yisrael Belsky, accept this identification based on strong scientific and textual evidence. However, other authorities, including Rav Joseph B. Soloveitchik, are reluctant to endorse the use of tekheiles without an unbroken tradition (Nefesh Ha-Rav, pp. 53-54). They argue that empirical identification, no matter how convincing, does not carry the same halakhic weight as tradition. This is particularly true when arguing based on evidence that is largely circumstantial, attempting to correlate contemporary facts to ancient descriptions. While AI may offer valuable tools to support the investigation of halakhic items, its conclusions will be bound by the limits of evidence-based research.

For those who believe that the location of the altar can be determined empirically, AI may help decisively show us how and where to build it. But for those who maintain that only prophecy can permit the rebuilding, no algorithm can replace a prophet. According to this view, AI might help us look back with clarity, but only prophecy can help us move forward with certainty. In the end, AI may clarify what we know, but it cannot authorize what we do.

Seeing is believing, until that too is taken away from us. The rapid development of artificial intelligence (AI) has produced powerful new tools, among them the ability to create “deepfakes,” which are highly convincing but entirely fake images, audio recordings or videos of real people. A politician may appear to endorse a controversial policy, a rabbi might seem to utter heretical ideas or a teacher could be seen acting inappropriately but it is all through manipulated media. These synthetic creations are increasingly difficult to distinguish from reality and pose serious ethical and halakhic questions, and paradoxically might also offer a solution to a difficult social problem.

I. Libelous Content

Let us begin with the most obvious victim: the individual whose likeness is misused. If a deepfake attributes scandalous or false words to a person, this constitutes classic hotza’as shem ra, false speech. Rav Yisrael Kagan (20th cen., Russia) quotes Onkelos (Lev. 19:16), who translates “Do not be a talebearer” as “lo seichol kurtzin.” Rashi (ad loc.) explains that this refers to the way gossippers motion with their eyes. Even indirect gossip, even mere motioning without any words, qualifies as forbidden lashon ha-ra (Chafetz Chaim 1:1:8 n. 13). As it says in Mishlei (Prov. 6:12-13), “A base person… winks with his eyes, scrapes with his feet, points with his fingers.” Lashon ha-ra does not have to be actual words. Posting an image or a video accusing someone of a misdeed constitutes forbidden speech, even if it is not verbal.

Hurtful content goes further. Posting a video that hurts someone’s feelings or damages someone’s reputation is a form of attack. This is biblically prohibited as ona’as devarim, hurtful and damaging speech (Bava Metzi’a 58b). You may think that this only applies to a verbal insult or attack–after all, “devarim” means words. However, the Torah (Lev. 25:17) merely says “lo sonu” and does not differentiate between methods of delivering this harm. When someone is portrayed doing or saying something inappropriate through a deepfake video, it violates their reputation and causes social, professional and emotional harm. Even if the victim is ultimately vindicated, the damage may already be irreversible.

II. Theft of a Likeness

There is also the question of image rights. Do I have halakhic control over my likeness? Rav Yosef Chaim Sonnenfeld (20th cen., Israel) did not object to people being photographed without consent, arguing that it causes no tangible damage (Salmas Chaim, no. 475). In contrast, Rav Menashe Klein (20th cen., US) contends that unauthorized photography may be prohibited because a person’s image has commercial and personal value. He notes that celebrities sell their likenesses, and therefore, using someone’s image without permission can constitute a form of theft (Mishneh Halakhos 7:117). Rav Simcha Yonah Klein (cont., England) follows Rav Sonnenfeld’s view but adds an important caveat: even the lenient view forbids publicizing damaging images (Piskei Ha-Mishpat 363:14).

Rav Shlomo Aviner (cont., Israel) takes a more fundamental approach, invoking the verse, “ve-ahavta le-rei’akha kamokha” (Lev. 19:18), which means that you must treat others as you would wish to be treated (Shabbos 31a). If a person finds it offensive or degrading to have his picture used without consent, that alone may be grounds for objection (Piskei Shlomo, vol. 3, pp. 146-147).

A deepfake is not merely a picture but a full impersonation, crafted to deceive and often to defame. The halakhic violations here are compounded: falsehood, humiliation, possible theft and an affront to human dignity.

III. Misleading the Public

While the subject of the deepfake suffers personal harm, halakhah also recognizes the broader concern of deceiving the public. The Talmud (Chullin 94a) prohibits geneivas da’as, deception. Rashi explains that misleading someone, even without causing financial loss, is forbidden. Even pretending to do someone a favor, when you really did it for personal reasons, is forbidden. Rambam (Mishneh Torah, Hilkhos De’os 2:6) offers other examples of deceptions that fall under this prohibition. The Torah says, “mi-dvar sheker tirchak, stay far away from falsehood” (Ex. 23:7). We are forbidden not just to lie but even to approach the point of generating falsehood.

Beyond personal injury, deepfakes pose broader communal dangers. When a large audience sees a fabricated video of a community leader making a controversial statement or a public figure endorsing a problematic idea, they are victims of falsehood. The consequences can reverberate in practical decisions and personal attitudes. In the past, we have seen disinformation sway public debates. Deepfakes could do the same on a much larger scale.

The implications are troubling. The erosion of trust in public communication affects not only our perception of individuals but also our confidence in halakhic, political, educational and journalistic institutions. When people begin to question the authenticity of everything they see, even genuine content becomes suspect. That social instability undermines public discourse. While a defamed individual suffers acutely, the collective deception may be more insidious. It destabilizes communal norms and damages the integrity of public discourse.

IV. A Silver Lining

Despite these serious concerns, the deepfake phenomenon may inadvertently contain a silver lining. Deepfakes might ultimately prompt society to think more critically about information. As people grow aware that images, videos and audio can be manipulated, they may become more discerning: Who recorded this? Where was it published? Has it been verified?

This shift could lead to a decline in so-called citizen journalism, i.e. social media reporting, and a resurgence of serious journalism. Media outlets that engage in careful fact-checking, apply editorial oversight and maintain transparent sourcing may gain renewed relevance. In an environment dominated by outrage and virality, credibility may once again become a competitive advantage.

For years, media has been pulled by the gravity of outrage and speed, chasing clicks and engagement at the expense of accuracy. The deepfake crisis may reverse that trend by incentivizing credibility over virality. In doing so, it would return media to its original mission: not to provoke, but to inform; not to inflame, but to illuminate.

Artificial Intelligence (AI) is increasingly shaping our world, raising new and complex questions. One such issue is whether it is permissible to use AI tools that analyze email and message content in order analyze trends without obtaining individual information. Consider an AI tool that reads people’s direct messages to analyze business trends and advises you on investment strategies. This tool can be perfectly legal but halakhah might be stricter. Are we allowed to benefit from AI analysis of private emails and messages?

I. The Cherem of Rabbeinu Gershom

Rabbeinu Gershom (11th cen., Germany) gathered leading rabbis together to issue a cherem—a communal ban—against a number of practices. Historians debate his specific role and the exact content and list of practices that fell under this ban, as opposed to later bans (Chakmei Ashkenaz Ha-Rishonim, p. 132ff). Regardless, there is a long-standing ban against opening and reading someone else’s mail that is widely attributed to Rabbeinu Gershom. The original context was physical letters, but authorities have extend its application to modern communication, including emails and text messages.

Rav Simchah Yonah Klein (cont., England) writes that the cherem applies even to hidden video cameras, i.e. nanny cams (Piskei Ha-Mishpat 154:7). Similarly, Rav Shaul Alter, in his recent book Software and Artificial Intelligence in Halacha, writes that “the cherem is not limited specifically to reading a letter, but to consuming any information that was not intended for public knowledge” (p. 334). It seems clear that emails and private messages fall under the cherem, as well.

Rav Chaim Palaggi (19th cen., Turkey) asks what rationale lies beneath this cherem? Even if the act is independently prohibited, Rabbenu Gershom and his colleagues may have enacted a cherem to reinforce the prohibition. Rav Palaggi suggests that it might be the obligation to love your fellow like yourself, or rather to not do to others what you would not want done to you (Lev. 19:18; Shabbos 31a). It might be a variant of gossip-mongering (Lev. 19:16). Alternatively, it might be a form of deception, geneivas da’as, or of using someone’s property without permission, which constitutes theft (Responsa Chikekei Lev 1:49). The distinctions between these different rationales will be crucial for evaluating the permissibility of AI that accesses private communications.

II. Rationales and Implications

Some authorities understand the prohibition as a form of theft by accessing someone else’s property, his information. Rav Chaim Shabesai (17th cen., Greece) writes that opening someone else’s letter is like borrowing something without permission, which is a form of stealing (Responsa Toras Chaim 3:47). In his recently republished book, The Practical Torah, Rav Michael Taubes writes that according to this view, “such action is forbidden even for the purpose of a mitzvah, particularly since one can cause damage to another by revealing his secrets” (p. 297).

According to this approach, if an AI accesses private email data without the owner’s consent, then the developers or users of the AI tool may be considered to have committed a form of theft. Even if no human sees the content, the act of extracting value from someone’s private correspondence without permission would still fall under the category of theft.

On the other hand, several authorities understand the core of the cherem not as theft, but as a violation of privacy. Rav Ya’akov Chagiz (17th cen., Morocco) briefly says that opening someone else’s mail violates the prohibition against gossip-mongering (Halakhos Ketanos 1:276). As mentioned above, Rav Chaim Palaggi suggests that this might be a violation of loving your fellow as yourself. However, both of those reasons would permit opening someone else’s mail when there is a mitzvah need to do so.

According to this approach, the violation is not the act of accessing information per se, but the act of disclosure. Meaning, the problem is violating the privacy of the individual. If so, an AI system that ingests an email but does not reveal the content, only extracting patterns or offering aggregate insights, does not violate the individual’s privacy. The secret remains hidden; no human sees it, no one exposes it.

For example, if an AI reads a team’s emails and determines that a particular merger will go through, but does not expose the personal content of any individual message, it might be functionally equivalent to a locked diary read by no one. The information is accessed, but the individual’s personal privacy remains intact.

III. Additional Considerations

According to our analysis, if reading someone’s mail is forbidden because of gossip-mongering, an AI would be allowed to analyze private correspondence and provide anonymized and aggregated analyses which could be used for business purposes. However, if it is prohibited because of theft, then even anonymizing and aggregating the information would still be forbidden.

However, Rav Yosef Shalom Elyashiv (21st cen., Israel) is reported as modifying Rav Palaggi’s analysis in a significant way. Rav Elyashiv’s son-in-law, Rav Yitzchak Zilberstein, presented a related question to him. Rav Elyashiv replied that opening someone’s mail can only constitute theft if the information is used to interfere with the victim’s plans. But if the victim does not suffer any loss, he has not been subjected to theft (Chashukei Chemed, Yoma, p. 55). If that is the case, then we would be allowed to invest based on advice legally extracted by AI from private emails and correspondence as long as it does not cause any damage to other investors.

I. AI and Video Supervision

As artificial intelligence (AI) makes its way into the workforce, it is worth asking whether it can serve as a mashgiach, a kosher inspector. While in the future there may be AI robots that can visit restaurants and factories, we do not have to look ahead to consider this question. Already today, many factories and restaurants are equipped with video cameras that enable remote supervision. During Covid, the use of such surveillance became even more widespread.

One problem with video surveillance is that it requires someone to watch the camera feeds, which can become a significant challenge as kosher supervision cameras proliferate. AI may offer a solution to this problem. Can AI be trusted to inspect the camera feeds and certify that nothing improper occurs at the establishments under supervision?

AI technology paired with live video feeds has already become a valuable tool in general surveillance. It monitors activity, detects anomalies and provides real-time data on temperature, movement and workflow. In fact, it can also monitor other electronic information about food production, such as ingredient ordering and delivery schedules. AI can even reduce human error or alert to suspicious behavior. However, kosher inspection involves more than monitoring.

A mashgiach must be aware of the broader context: how ingredients are sourced, how utensils are maintained and how personnel follow halakhic restrictions. Nothing can replace an in-person inspection. A simple conversation with employees reveals a lot about the professionalism of the operation, unexpected changes that have happened or are about to happen, and general information about personnel and processes. Additionally, cameras can be manipulated and visual access may not capture essential details.

Be that as it may, there still is room for AI to serve as the video mashgiach that works alongside human inspectors. Is AI halakhically trusted for this work?

II. Mashgiach: Testimony or Trust?

Very often, a mashgiach is not permanently stationed at a factory or restaurant. He comes and goes periodically (yotzei ve-nikhnas), conducting spot checks. The halakhic mechanism that allows yotzei ve-nikhnas is the concept of mirsas, fear. Namely, the workers do not violate proper procedures because they fear being caught by the mashgiach on a spot check or on a video stream. This fear serves as a deterrent and ensures compliance.

When a mashgiach affirms a factory’s procedures, is he testifying to the kosher status of the food? If he has not witnessed the food’s production, it is not clear whether he can testify to that. Perhaps the concept of mirsas enables the mashgiach to testify even when he does not see the entire production. Or, alternatively, it eliminates the need for testimony altogether, allowing the mashgiach to simply report facts on the ground. Put differently, is the mashgiach testifying about the food or is he serving as an agent of the supervising rabbi who testifies about the food? As long as the mashgiach verifies the food production to the satisfaction of the supervising rabbi’s halakhic requirements, the supervising rabbi – whose name appears on the kosher certificate – can testify that the food is kosher. If so, AI can serve as a mashgiach.

On the other hand, if a mashgiach is considered a witness, then the laws of testimony for forbidden foods apply. Only an observant Jewish adult can serve in such a capacity. A gentile, a child or someone lacking halakhic competence is disqualified. AI, which lacks personhood and obligation in mitzvos, would certainly be excluded on these grounds.

III. Women and AI

The question of AI as a mashgiach can be informed by the halakhic discussion about whether women can serve in kosher supervision. Rav Moshe Feinstein (20th cen., US) argues that a woman may serve as a kosher supervisor (Igros Moshe, Yoreh De’ah 2:44). Rav Feinstein addresses two concerns: 1) is a woman’s testimony accepted in complex kosher matters? To this, he answers yes. 2) Can a woman serve in a communal position of authority? To this, Rav Feinstein answers no but suggests that she appoint a supervising rabbi to formally issue the kosher certificate while she serves as the kosher inspector.

Can this model work for AI also? It seems not, because Rav Feinstein still requires the woman inspector to testify to the supervising rabbi about the food’s kosher status. However, Rav Feinstein opens the door to a different framework. In the directly preceding responsum (op cit., 43), he writes that in many practical areas of kashrus, we rely not on formal testimony but on ne’emanus – the halakhic assumption of trustworthiness. When someone is known to be honest and trustworthy, halakhah allows reliance on their word even without the structure of testimony (Igros Moshe, Yoreh De’ah 2:43). If so, then maybe even AI can be relied upon even if the mashgiach testifies to the food’s kosher status. A well-trained AI, thoroughly tested and validated, may be even more trustworthy than a human because it has no personal interests or distractions that might compromise its judgment.

If so, it seems that when it comes to monitoring video feeds, AI can serve as a kosher inspector that provides important information for the consideration of the supervising rabbi. This is true assuming that the AI has been tested and verified as an accurate inspector of videos. This also assumes that a separate inspector makes site visits to review other pertinent information.

Potentially, AI is a powerful tool of kosher supervision. It can inspect important parts of the kosher food production process. However, currently it cannot replace the site visit and merely supplements the visit with the monitoring of electronic surveillance.