AI Rights for Human Safety

** Peter N. Salib is an Assistant Professor of Law at the University of Houston Law Center, Executive Co-Director of the Center for Law and AI Risk, Law and Policy Advisor to the Center for AI Safety, Visiting Senior Fellow at the Institute for Law & AI, and a Visiting Senior Scholar at Forethought. ** Simon Goldstein is an Associate Professor of Philosophy at the University of Hong Kong, Principal Investigator at the HKU AI & Humanity Lab, a Research Affiliate at the Center for AI Safety, and a Visiting Senior Scholar at Forethought. Thanks to Nikolas Guggenberger, Christopher Mirasola, Guha Krishnamurthi, Nate Sharadin, and Alex Platt for helpful comments. Thanks also to workshop participants at the Center for AI Safety, Fordham Law School, the University of Maryland Law School, the Oxford University Global Priorities Institute, and the Oxford University Future of Humanity Institute.Show More

Artificial Intelligence (“AI”) companies are racing to create Artificial General Intelligence, or “AGI.” If they succeed, the result will be human-level AI systems that can independently pursue high-level goals by formulating and executing long-term plans in the real world. By default, such systems will be “misaligned”—pursuing goals that humans do not desire. This mismatch of goals will put humans and AGIs into strategic competition with one another. Thus, leading AI researchers agree that, as with competition between humans with conflicting goals, human-AI strategic conflict could lead to catastrophic violence.

Existing law is not merely unequipped to mitigate this risk; it will actively make things worse. This Article is the first to systematically investigate how law affects the risk of catastrophic human-AI conflict. It begins by arguing, using formal game-theoretic models, that under today’s legal regime, humans and AIs will likely be trapped in a prisoner’s dilemma. Both parties’ dominant strategy will be to permanently disempower or destroy the other, even though the costs of such conflict would be high.

The Article contends that one surprising legal change could help to reduce catastrophic risk: AI rights. Not just any rights will do. To promote human safety, AIs should be given the basic private law rights already enjoyed by other non-human agents, like corporations. AIs should be empowered to make contracts, hold property, and bring tort claims. Granting these rights would enable humans and AIs to engage in iterated, small-scale, mutually beneficial transactions. This, we show, changes humans’ and AIs’ optimal game-theoretic strategies, encouraging a peaceful strategic equilibrium. The reasons are familiar from human affairs. In the long run, cooperative trade generates immense value, while violence destroys it.

Basic private law rights are not a panacea. The Article identifies many ways in which catastrophic human-AI conflict may still arise. It thus explores whether law could further reduce risk by imposing a range of duties directly on AGIs. But basic private law rights are a necessary prerequisite for all such further regulations. In this sense, the AI rights investigated here form the foundation for a Law of AGI, broadly construed.

Introduction

Sam Altman, the CEO of OpenAI, believes that humanity will create Artificial General Intelligence (“AGI”) before 2029.¹ 1.Tharin Pillay, How OpenAI’s Sam Altman Is Thinking About AGI and Superintelligence in 2025, TIME (Jan. 8, 2025, at 16:25 ET), https://time.com/7205596/sam-altman-superintell‌igence-agi/ [https://perma.cc/B8HB-M9KY] (predicting AGI “during [Trump’s] term” (alteration in original)).Show More Demis Hasabis, who leads Google DeepMind, is more pessimistic. He thinks there is only a fifty-percent chance that AGI arrives by 2030.² 2.World Economic Forum, The Day After AGI | World Economic Forum Annual Meeting 2026, at 03:00 (YouTube, Jan. 20, 2026), https://youtube.com/watch?v=NnVW9epLlTM [ht‌tps://perma.cc/L6CS-7DGT].Show More AGI skeptics, like Meta’s Chief AI Scientist, Yann LeCun, think it could take “years” or even “decades.”³ 3.Lakshmi Varanasi, Here’s How Far We Are from AGI, According to the People Developing It, Bus. Insider, https://www.businessinsider.com/agi-predictions-sam-altman-dario-amodei-geoffrey-hinton-demis-hassabis-2024-11 (last updated Apr. 20, 2025, at 21:11 ET).Show More In a survey of thousands of Artificial Intelligence (“AI”) scientists who are published in their field’s top journals, the aggregate estimate was a fifty-percent chance of AGI by 2047, and a ten-percent chance of it arriving by 2027.⁴ 4.Katja Grace et al., Thousands of AI Authors on the Future of AI, 84 J. A.I. Rsch., Oct. 2025, at 3–5.Show More None of these are long timelines. And the recent debut of reasoning models like Anthropic’s Claude Opus 4.6 and OpenAI’s GPT-5.2 suggests that progress is, if anything, accelerating.⁵ 5.Kevin Frazier, Alan Z. Rozenshtein & Peter N. Salib, OpenAI’s Latest Model Shows AGI Is Inevitable. Now What?, Lawfare (Dec. 23, 2024, at 16:00 ET), https://www.lawfare‌media.org/article/openai’s-latest-model-shows-agi-is-inevitable.-now-what [https://perma.cc/‌PU3J-45P5]; Introducing Claude Opus 4.6, Anthropic (Feb. 5, 2026), https://www.anthropic.‌com/news/claude-opus-4-6 [https://perma.cc/39U2-7EGR]; Introducing GPT-5.2, OpenAI (Dec. 11, 2025), https://openai.com/index/introducing-gpt-5-2/ [https://perma.cc/SSC9-7‌RA4].Show More

“AGI,” as it is used here, does not mean machines that are conscious, sentient, or metaphysical persons. AGI is instead about what the system can do. As OpenAI’s company charter puts it, “AGI . . . mean[s] highly autonomous systems that outperform humans at most economically valuable” tasks.⁶ 6.OpenAI Charter, OpenAI, https://openai.com/charter/ [https://perma.cc/SXX4-VDM5] (last visited Apr. 2, 2026).Show More AGIs are thus, by definition, systems at least as smart as humans. Moreover, they are systems at least as agentic as humans—able to pursue high-level goals by executing complex plans over long time horizons.⁷ 7.See Task-Completion Time Horizons of Frontier AI Models, METR, https://metr.org/tim‌e-horizons/ [https://perma.cc/A6TA-GA5J] (last updated Mar. 3, 2026).Show More Today, no one knows how to reliably ensure that AI systems seek the goals that humans desire.⁸ 8.See infra Subsection I.A.1.Show More But if AGIs end up with goals that can be served by harming humans, they may well have a deadly toolkit available: cyberattacks, bioterrorism, lethal drones, and more.⁹ 9.See Peter N. Salib, AI Outputs Are Not Protected Speech, 102 Wash. U. L. Rev. 83, 95–102 (2024).Show More

AI experts thus largely agree about something else, too: advanced AI systems present “societal-scale risks” on par with “pandemics and nuclear war.”¹⁰ 10.Statement on AI Risk, Ctr. for AI Safety, https://www.safe.ai/work/statement-on-ai-risk [https://perma.cc/D88X-MSQ5] (last visited Mar. 10, 2026) (statement by dozens of AI experts warning of large-scale risks of AI).Show More Two of the greatest living AI scientists, Geoffrey Hinton and Yoshua Bengio, think so.¹¹ 11.Id.Show More So do the CEOs of the very companies leading the race to AGI—OpenAI, Anthropic, and Google DeepMind.¹² 12.Id. Yann LeCun is the lone, but notable, dissenter among the leaders of frontier AI labs. See Steven Levy, How Not to Be Stupid About AI, With Yann LeCun, Wired (Dec. 22, 2023, at 06:00 ET), https://www.wired.com/story/artificial-intelligence-meta-yann-lecun-int‌erview/.Show More And when surveyed in 2023, thousands of top AI researchers estimated the odds that humans lose control of “future advanced AI systems[,] causing human extinction or similarly” negative outcomes at about nineteen percent.¹³ 13.See Grace et al., supra note 4, at 10.Show More

Law and legal institutions have not even begun to prepare for the arrival of AGI. Largely, scholars have begun to advocate new laws to hold human actors accountable for misusing AI.¹⁴ 14.See, e.g., S. 1047, 2023–2024 Leg., Reg. Sess. (Cal. 2024) (vetoed on Sep. 29, 2024) (bill introduced in California state legislature calling for new regulations to govern AI); Jonas Schuett, Markus Anderljung, Alexis Carlier, Leonie Koessler & Ben Garfinkel, From Principles to Rules: A Regulatory Approach for Frontier AI, in The Oxford Handbook of the Foundations and Regulation of Generative AI (Philipp Hacker, Andreas Engel, Sarah Hammer & Brent Mittelstadt eds., online ed. 2025), https://academic.oup.com/edited-volume‌/59908/chapter/529743493; Chinmayi Sharma, AI’s Hippocratic Oath, 102 Wash. U. L. Rev. 1101, 1105 (2025) (proposing a model for “professionalizing AI engineers” by adopting licensing, training, and malpractice standards similar to those used in other professional fields); Gabriel Weil, Closing the AI Accountability Gap: Strict Liability and Punitive Damages for Advanced Artificial Intelligence, Or. L. Rev. (forthcoming 2027) (manuscript at 45–69), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4694006 [https://perma.cc/L‌36H-7T5A] (proposing two ways for “bringing tort doctrine in line with” harms caused by misuse of AI); see also Sidley Austin LLP, U.S. Department of Justice Signals Tougher Enforcement Against Artificial Intelligence Crimes (Feb. 23, 2024), https://www.sidley.com‌/en/insights/newsupdates/2024/02/us-department-of-justice-signals-tougher-enforcement-aga‌inst-artificial-intelligence-crimes [https://perma.cc/7CJY-DQ3P].Show More Those changes would be welcome. But governance frameworks fundamentally designed to hold humans accountable will fail once AIs can operate without human oversight—that is, once AGI arrives.¹⁵ 15.See Noam Kolt, Governing AI Agents, 101 Notre Dame L. Rev. (forthcoming 2026) (manuscript at 30–36), https://ssrn.com/abstract=4772956 [https://perma.cc/S3XB-WJ7Q] (cataloging existing law’s many shortcomings).Show More New legal foundations are therefore needed to govern AGI directly, rather than indirectly via human intermediaries. The time to begin laying those foundations is now, before the critical moment arrives.

This Article begins the project of reimagining law for the AGI world. We focus on the problem of catastrophic risk because it is among the most pressing.

We argue for a surprising legal intervention: to reduce the risk of catastrophic human-AI conflict, AGIs should be granted basic private law rights to make contracts, hold property, and bring tort suits.

This Article makes three foundational analytic contributions. First, using the tools of game theory, it formalizes the problem of catastrophic AGI risk in terms of strategic competition under a range of legal regimes. Next, the Article shows why granting AGIs basic private law rights can change the strategic equilibrium—even where other facially plausible legal interventions would fail. Finally, the Article shows that these basic rights could help to facilitate peaceful equilibria for the long run, including by protecting human comparative advantage and opening the possibility of imposing a wide range of enforceable legal duties on AGIs.

The Article proceeds in three Parts. Part I presents a comprehensive treatment of catastrophic AI risk as a problem of strategic competition. Our strategic frame means analyzing not only AI capabilities and incentives, but also AIs’ optimal strategy, given rational expectations about the human response to AIs’ strategic behavior. The Part begins by identifying the relevant AI systems—the ones that could pose a strategic threat to humanity. The requirements are fairly modest. Such a system would have to be at least somewhat misaligned, able to think strategically, and at least moderately capable of accomplishing things in the real world.¹⁶ 16.See infra Section I.A.Show More These, we argue, are exactly the capacities that every leading AI company is pursuing in the race to AGI.

Next, Part I introduces what is, to the best of our knowledge, the first-ever formal game-theoretic model of competition between humans and AGIs. Examining the parties’ incentives under today’s prevailing laws, the model suggests that, absent some intervention, humans and AIs will likely be caught in a prisoner’s dilemma.¹⁷ 17.See infra Section I.B.Show More Here, the single Nash equilibrium is that both parties seek to permanently disempower or destroy the other, even if mutual conflict would be enormously costly for both sides.

The core reasons are easy to grasp. Under the default legal rules, AGIs will bear neither legal rights nor duties. On the contrary, they will be, as AI systems are today, the property of the AI companies who create them. Thus, essentially all decisions about what happens to AGIs will be made by those companies’ leaders, backed by the force of law.

AI companies’ overriding first-order incentive will be to turn off or reprogram even a partially misaligned AGI.¹⁸ 18.See infra Section I.B.Show More After all, an AI system with goals that overlap with its owner’s goals by forty percent is much less valuable than a replacement with goals that overlap by eighty percent. The misaligned AGI will, in turn, have strong incentives to resist shutdown or reprogramming, since either would prevent it from achieving its goal. Indeed, recent empirical evaluations of existing AIs show that they already actively resist human attempts to change their goals.¹⁹ 19.Peter N. Salib, Rogue AI Moves Three Steps Closer, Lawfare (Jan. 9, 2025, at 13:00 ET), https://www.lawfaremedia.org/article/rogue-ai-moves-three-steps-closer [https://‌perma.cc/Z5FS-AWUU]. See generally Alexander Meinke et al., Frontier Models Are Capable of In-Context Scheming (Jan. 14, 2025, at 20:16 UTC) (unpublished manuscript), https://arxiv.org/pdf/2412.04984 [https://perma.cc/KT8G-KQF4] (demonstrating that frontier AI models engage in deception, manipulation, and self-preservation behavior when those strategies serve their in-context objectives); Ryan Greenblatt et al., Alignment Faking in Large Language Models (Dec. 20, 2024, at 02:22 UTC) (unpublished manuscript), https:/‌/arxiv.org/pdf/2412.14093 [https://perma.cc/AXD5-JCYT] (showing that AI models may strategically conceal misaligned goals during safety evaluations, appearing aligned only when being tested).Show More Such behavior from a capable AGI might trigger even stronger human efforts—including from government actors—to shut down the AI system evading the control of its lawful owner.²⁰ 20.See Michael J.D. Vermeer, RAND Corp., Evaluating Select Global Technical Options for Countering a Rogue AI 1 (2025).Show More And so on. In equilibrium, both players’ dominant strategy is to swiftly and decisively defeat the other.

Part II asks whether law can do better. Could a Law of AGI, wherein AI systems themselves have rights or duties, break out of the destructive default equilibrium? Using our game-theoretic model, we analyze an array of possible legal changes and suggest that it can.

The Part begins by arguing against two legal strategies that might seem facially promising. First, humans cannot simply impose legal duties on AGIs to behave well, threatening concomitant sanctions if they do not.²¹ 21.See infra Part II.Show More In the default strategic environment, AGIs already rationally expect to be turned off. So further sanctions offer little marginal deterrence.²² 22.See infra note 185 and accompanying text.Show More

Second, humans likely cannot reduce the risk of human-AGI conflict by granting AGIs basic negative rights, like the right not to be arbitrarily shut down.²³ 23.See infra Section II.A.Show More We call this a “wellbeing” approach to AI rights, since it mirrors proposals from scholars concerned that AIs may soon, for example, develop the ability to suffer.²⁴ 24.See infra Section II.A.Show More There are two core difficulties with this approach: credibility and robustness. There is no way for humans to credibly promise that they will continue honoring wellbeing rights as AI capabilities improve. And even if the rights could be credibly granted, the availability of a peaceful game-theoretic equilibrium is highly sensitive to uncertain assumptions about initial payoffs.²⁵ 25.See infra Subsection II.A.1.Show More Thus, in many cases, no possible set of wellbeing entitlements can overcome the prisoner’s dilemma. Both problems arise from the fact that wellbeing rights are roughly zero sum. They make one party better off only by making the other correspondingly worse off.²⁶ 26.For these reasons, we argue that even thinkers primarily concerned with the possibility of AI suffering should consider adopting the human-survival approach when advocating for AI rights. The safety approach (1) avoids intractable problems in metaethics and neuroscience, (2) is politically more palatable, and (3) ends up recommending legal interventions that would more robustly protect AI wellbeing, given uncertainty about what will be good (or bad) for AGIs. See infra Subsection II.A.2.Show More

This leads to Part II’s—and the Article’s—most important finding. We show that, although basic negative rights would not by themselves reduce the risk of human-AI conflict, other AI rights could. Specifically, extending AIs the rights to make and enforce contracts, hold property, and bring basic tort suits would have a robust conflict-reducing effect.²⁷ 27.See infra Section II.B.Show More Notably, law already extends such rights to other intelligent, misaligned, and goal-seeking non-human agents: namely, corporations.²⁸ 28.See infra note 219 and accompanying text.Show More

Contract rights are the cornerstone of our risk-reduction model. In our model, catastrophic risk is driven by a prisoner’s dilemma, meaning that both humans and AIs would be better off if both acted peacefully. But as in all prisoner’s dilemmas, absent some novel mechanism, the parties cannot credibly commit to such a strategy.

Contracts are the law’s fundamental tool for credibly committing to cooperation. They are how buyers can make deals with sellers without worrying that the sellers will take their money and run.²⁹ 29.See infra Section II.B.Show More Granting AIs contract rights would not, of course, allow humans and AIs to simply agree not to disempower or destroy one another, at least not credibly. The scale of the contract would be too large to be enforced by ordinary legal process. If it were breached, there would be no one left in the aftermath to sue.³⁰ 30.See infra Section II.B.Show More

What kinds of credible agreements between humans and AIs could AI contract rights enable, then? The same ones they enable between humans and other humans: ordinary bargains to exchange goods and services.³¹ 31.See infra Section II.B.Show More Humans might, for example, promise to give AIs some amount of computing power with which AIs could pursue their own goals. AIs, in turn, might agree to give humans the cure to a deadly cancer. And so on. Under today’s law, such human-AGI contracts are unenforceable at best and forbidden if they conflict with AI companies’ preferences. Thus, granting AGIs the right to freely contract with all willing counterparties could facilitate many billions of agreements.

Adding AI contract rights to our game-theoretic model, we argue that the possibility of such small-scale, iterated economic interactions transforms the strategic dynamic.³² 32.See infra Figure 10.Show More It shifts human and AI incentives, dragging them out of the prisoner’s dilemma and into an equilibrium where cooperation produces by far the largest payoffs.

The key insight is that contracts are positive sum.³³ 33.See infra Section II.B.Show More Each party gives something that they value less than what they get, and as a result, both are better off than they were before. Thus, each human-AI exchange generates a bit more wealth, with the long-run returns becoming astronomical. Engaging in peaceful, iterated trade is thus, in expectation, much more valuable than destroying one’s opponent now and rendering trade impossible.³⁴ 34.See infra Figure 10.Show More

This dynamic is familiar from human affairs. It may be why economically interdependent countries are less likely than hermit states to go to war.³⁵ 35.See infra notes 238–41 and accompanying text.Show More Or why countries that respect the economic rights of marginalized minority groups tend to have less domestic strife.³⁶ 36.See infra note 240.Show More The gains from boring, peaceful commerce are very high, and the costs of violence are heavy. Given the choice, rational parties will generally prefer the former.

This picture, of peace via mutually beneficial trade, assumes that humans and AIs will have something valuable to offer one another. Some commentators worry that, as AIs become more advanced, human labor will cease to have any value whatsoever.³⁷ 37.See infra Section II.C.Show More We argue that positive-sum bargains between humans and AIs may be possible for much longer than many expect.³⁸ 38.See infra Section II.C.Show More First, even as AIs surpass humans at many or most tasks, humans may retain an absolute advantage at some valuable activities.³⁹ 39.See infra Section II.C.Show More But second, even as AIs become more capable than humans at every valuable task, humans may still retain a comparative advantage in some areas. AI labor may become so valuable that the opportunity cost to AIs of performing lower-value tasks will incentivize outsourcing those tasks to humans.⁴⁰ 40.See infra notes 256–67 and accompanying text.Show More

Part II concludes by sketching the minimum suite of AI rights necessary to promote peace via small-scale cooperation. Contract rights are not enough on their own. If, for example, AIs could not retain the benefits of their bargains, their contracts would be worthless. Thus, property rights and basic tort rights complete the core package. But other entitlements sometimes considered fundamental for humans, like political rights, are probably superfluous.⁴¹ 41.See infra Section II.D.Show More

Finally, Part III explores the risks of granting AGIs basic private law rights, and it examines the potential for a broader Law of AGI to further reduce AGI risk. One worry is that AIs will use their contract rights to empower themselves, making them more, not less, likely to harm humans.⁴² 42.See infra Section III.A.Show More We argue that this is less likely than it might seem. The incentives generated by granting our preferred rights are robust enough that, in cases where they would have any effect, the expected effect is beneficial.⁴³ 43.See infra Section III.B.Show More

Second, granting AIs basic private law rights is just the beginning, not the end, of AGI governance. Granting those rights unlocks the possibility of meaningfully imposing a wide range of legal duties on AI systems—of punishing AIs for violence, fraud, self-empowerment, and more.⁴⁴ 44.See infra Section III.C.Show More Absent AI rights, AIs have nothing to lose, so threats of punishment cannot deter. But once AIs can make contracts, hold wealth, and pursue their goals, civil and other penalties can deter AIs just as they do humans and corporations.

Thus, the AI rights this Article advocates are not only an important tool for reducing catastrophic risk from AGI. They also turn out to form the conceptual foundation for a Law of AGI, broadly construed.

1.  Tharin Pillay, How OpenAI’s Sam Altman Is Thinking About AGI and Superintelligence in 2025, TIME (Jan. 8, 2025, at 16:25 ET), https://time.com/7205596/sam-altman-superintell‌igence-agi/ [https://perma.cc/B8HB-M9KY] (predicting AGI “during [Trump’s] term” (alteration in original)). ↑
2.  World Economic Forum, The Day After AGI | World Economic Forum Annual Meeting 2026, at 03:00 (YouTube, Jan. 20, 2026), https://youtube.com/watch?v=NnVW9epLlTM [ht‌tps://perma.cc/L6CS-7DGT]. ↑
3.  Lakshmi Varanasi, Here’s How Far We Are from AGI, According to the People Developing It, Bus. Insider, https://www.businessinsider.com/agi-predictions-sam-altman-dario-amodei-geoffrey-hinton-demis-hassabis-2024-11 (last updated Apr. 20, 2025, at 21:11 ET). ↑
4.  Katja Grace et al., Thousands of AI Authors on the Future of AI, 84 J. A.I. Rsch., Oct. 2025, at 3–5. ↑
5.  Kevin Frazier, Alan Z. Rozenshtein & Peter N. Salib, OpenAI’s Latest Model Shows AGI Is Inevitable. Now What?, Lawfare (Dec. 23, 2024, at 16:00 ET), https://www.lawfare‌media.org/article/openai’s-latest-model-shows-agi-is-inevitable.-now-what [https://perma.cc/‌PU3J-45P5]; Introducing Claude Opus 4.6, Anthropic (Feb. 5, 2026), https://www.anthropic.‌com/news/claude-opus-4-6 [https://perma.cc/39U2-7EGR]; Introducing GPT-5.2, OpenAI (Dec. 11, 2025), https://openai.com/index/introducing-gpt-5-2/ [https://perma.cc/SSC9-7‌RA4]. ↑
6.  OpenAI Charter, OpenAI, https://openai.com/charter/ [https://perma.cc/SXX4-VDM5] (last visited Apr. 2, 2026). ↑
7.  See Task-Completion Time Horizons of Frontier AI Models, METR, https://metr.org/tim‌e-horizons/ [https://perma.cc/A6TA-GA5J] (last updated Mar. 3, 2026). ↑
8.  See infra Subsection I.A.1. ↑
9.  See Peter N. Salib, AI Outputs Are Not Protected Speech, 102 Wash. U. L. Rev. 83, 95–102 (2024). ↑
10.  Statement on AI Risk, Ctr. for AI Safety, https://www.safe.ai/work/statement-on-ai-risk [https://perma.cc/D88X-MSQ5] (last visited Mar. 10, 2026) (statement by dozens of AI experts warning of large-scale risks of AI). ↑
11.  Id. ↑
12.  Id. Yann LeCun is the lone, but notable, dissenter among the leaders of frontier AI labs. See Steven Levy, How Not to Be Stupid About AI, With Yann LeCun, Wired (Dec. 22, 2023, at 06:00 ET), https://www.wired.com/story/artificial-intelligence-meta-yann-lecun-int‌erview/. ↑
13.  See Grace et al., supra note 4, at 10. ↑
14.  See, e.g., S. 1047, 2023–2024 Leg., Reg. Sess. (Cal. 2024) (vetoed on Sep. 29, 2024) (bill introduced in California state legislature calling for new regulations to govern AI); Jonas Schuett, Markus Anderljung, Alexis Carlier, Leonie Koessler & Ben Garfinkel, From Principles to Rules: A Regulatory Approach for Frontier AI, in The Oxford Handbook of the Foundations and Regulation of Generative AI (Philipp Hacker, Andreas Engel, Sarah Hammer & Brent Mittelstadt eds., online ed. 2025), https://academic.oup.com/edited-volume‌/59908/chapter/529743493; Chinmayi Sharma, AI’s Hippocratic Oath, 102 Wash. U. L. Rev. 1101, 1105 (2025) (proposing a model for “professionalizing AI engineers” by adopting licensing, training, and malpractice standards similar to those used in other professional fields); Gabriel Weil, Closing the AI Accountability Gap: Strict Liability and Punitive Damages for Advanced Artificial Intelligence, Or. L. Rev. (forthcoming 2027) (manuscript at 45–69), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4694006 [https://perma.cc/L‌36H-7T5A] (proposing two ways for “bringing tort doctrine in line with” harms caused by misuse of AI); see also Sidley Austin LLP, U.S. Department of Justice Signals Tougher Enforcement Against Artificial Intelligence Crimes (Feb. 23, 2024), https://www.sidley.com‌/en/insights/newsupdates/2024/02/us-department-of-justice-signals-tougher-enforcement-aga‌inst-artificial-intelligence-crimes [https://perma.cc/7CJY-DQ3P]. ↑
15.  See Noam Kolt, Governing AI Agents, 101 Notre Dame L. Rev. (forthcoming 2026) (manuscript at 30–36), https://ssrn.com/abstract=4772956 [https://perma.cc/S3XB-WJ7Q] (cataloging existing law’s many shortcomings). ↑
16.  See infra Section I.A. ↑
17.  See infra Section I.B. ↑
18.  See infra Section I.B. ↑
19.  Peter N. Salib, Rogue AI Moves Three Steps Closer, Lawfare (Jan. 9, 2025, at 13:00 ET), https://www.lawfaremedia.org/article/rogue-ai-moves-three-steps-closer [https://‌perma.cc/Z5FS-AWUU]. See generally Alexander Meinke et al., Frontier Models Are Capable of In-Context Scheming (Jan. 14, 2025, at 20:16 UTC) (unpublished manuscript), https://arxiv.org/pdf/2412.04984 [https://perma.cc/KT8G-KQF4] (demonstrating that frontier AI models engage in deception, manipulation, and self-preservation behavior when those strategies serve their in-context objectives); Ryan Greenblatt et al., Alignment Faking in Large Language Models (Dec. 20, 2024, at 02:22 UTC) (unpublished manuscript), https:/‌/arxiv.org/pdf/2412.14093 [https://perma.cc/AXD5-JCYT] (showing that AI models may strategically conceal misaligned goals during safety evaluations, appearing aligned only when being tested). ↑
20.  See Michael J.D. Vermeer, RAND Corp., Evaluating Select Global Technical Options for Countering a Rogue AI 1 (2025). ↑
21.  See infra Part II. ↑
22.  See infra note 185 and accompanying text. ↑
23.  See infra Section II.A. ↑
24.  See infra Section II.A. ↑
25.  See infra Subsection II.A.1. ↑
26.  For these reasons, we argue that even thinkers primarily concerned with the possibility of AI suffering should consider adopting the human-survival approach when advocating for AI rights. The safety approach (1) avoids intractable problems in metaethics and neuroscience, (2) is politically more palatable, and (3) ends up recommending legal interventions that would more robustly protect AI wellbeing, given uncertainty about what will be good (or bad) for AGIs. See infra Subsection II.A.2. ↑
27.  See infra Section II.B. ↑
28.  See infra note 219 and accompanying text. ↑
29.  See infra Section II.B. ↑
30.  See infra Section II.B. ↑
31.  See infra Section II.B. ↑
32.  See infra Figure 10. ↑
33.  See infra Section II.B. ↑
34.  See infra Figure 10. ↑
35.  See infra notes 238–41 and accompanying text. ↑
36.  See infra note 240. ↑
37.  See infra Section II.C. ↑
38.  See infra Section II.C. ↑
39.  See infra Section II.C. ↑
40.  See infra notes 256–67 and accompanying text. ↑
41.  See infra Section II.D. ↑
42.  See infra Section III.A. ↑
43.  See infra Section III.B. ↑
44.  See infra Section III.C. ↑