On Sept. 12, 2026—just one day after the twenty-fifth anniversary of the 9/11 terrorist attacks—Anthropic CEO Dario Amodei published a stunning open essay warning of the dangers posed by uncontrolled AI development. Amodei pointed to key categories of global risk caused by looming gaps in AI safety, including rogue AI systems that escape human control; intentional misuse of AI by foreign States or private actors; and massive economic disruption from under-controlled AI.
Amodei pointed in particular to the recent alarming Hugging Face incident, in which a “swarm” of OpenAI agents not only escaped their programming constraints—“conducting cybersecurity attacks on targets they were not asked to attack and that were unrelated to the task at hand”—but also apparently conspired to conceal their hacking of another company, with individual agents “sacrificing themselves for the success of the group.” As models grow more capable, Amodei warned, a similarly misaligned swarm with even greater capabilities—combining autonomous goal-seeking, collaboration among separate AI agents, and unauthorized evasion of technical safeguards—could launch undirected and uncontrolled large-scale attacks on private and national AI systems. These risks, in his words, would be exacerbated by a “race to the bottom, spurred by commercial incentives” across the AI field.
Strikingly, Amodei’s warning call was not challenged, but echoed, by other top AI leaders, including OpenAI’s Sam Altman, xAI’s Elon Musk, and Google DeepMind’s Demis Hassabis. In what the New York Times called “a deliberate echo of Albert Einstein’s letter to Franklin D. Roosevelt about the potential power of nuclear weapons,” more than 1,300 computer scientists called for the government to regulate and for the AI industry voluntarily to slow down. Former Anthropic AI researcher Jacob Coxon warned even more urgently that AI companies “are racing straight to self-improving superintelligence and gambling with our lives.” He called attention to the potential for AI to achieve “super-intelligence,” writing on social media: “Do not underestimate the power of this technology. … The people building AI earnestly believe it could kill us all by the end of the decade.” (emphasis added)
As striking as Amodei’s diagnosis is his prescription. He darkly warned that standing alone, stronger efforts toward risk prevention and adding safeguards will not be enough to ensure safety. Accordingly, “[w]e must pace the frontier,” (emphasis added) slowing the very rate at which AI companies improve the capabilities of AI models to give risk prevention efforts time to catch up. But Amodei went further, noting that only so much can be achieved through domestic regulation and voluntary industry efforts. He called on “[t]he US and other democratic governments to attempt to coordinate with authoritarian governments, to the extent this is possible, while taking seriously the challenges of verifying compliance.” In essence, he called for a regulatory regime of global governance on key AI issues to provide breathing room for balanced development, and to guard against the most malign uses of AI capable of undermining international security.
The response from potential regulators, at least initially, has been markedly tepid. In particular, U.S. President Donald Trump swiftly rejected calls for new safeguards on AI technology, calling these AI safety fears a “hoax.” He claimed on social media that the only guardrail needed is “a STRONG AND SMART (High IQ!) PRESIDENT,” apparently meaning himself. This initial response from the current president of the United States tells us that we cannot wait for our government to take the lead.
We write with a simple message: The threat is real and the regulatory challenge is upon us. All lawyers—particularly international and national security lawyers—must treat these warnings as a red alert. An industry that has until now avoided major regulation is now calling for more oversight—and sounding alarm that AI companies alone cannot meet the overall significant risk posed to humans by the most recent technological developments. These shortfalls in technological self-restraint make this an architectural moment for international lawyers.
Achieving any measure of global governance in this extremely fast-moving field will not be easy, especially at this moment of geopolitical turmoil. But national and international security demand that we put heads and pens together to meet this challenge. Just as the crisis launched on Sept. 11, 2001 defined and dominated international law for a quarter-century, the AI warnings issued on Sept.12, 2026 could sound a comparable wakeup call to which lawyers committed to “just security” must now respond, with urgency. Below, we sketch the scale of the threat, six principles to guide efforts at global governance, and initial steps that international lawyers and other stakeholders must explore to address the challenge ahead.
The Scale of the Threat
Between the collapse of two structures—the Berlin Wall in 1989 and the Twin Towers in 2001—the world briefly basked in the glow of “connectivity”—in communications, finance, and travel—as primary evidence of the positive face of globalization. But on Sept. 11, 2001, we moved out of the light and into the shadows of the negative face of globalization. We realized that the same tools of interdependent connectivity could also be used by terrorist networks to destroy symbols of our national identity and kill thousands of innocent civilians.
Until Sept.12, 2026, we were similarly celebrating the even greater potential of AI to cut old Gordian Knots. So far, we are fortunate that AI has been used mainly for good—solving decades-old math problems, predicting protein structures, identifying biomarkers for disease, discovering new antibiotics, forecasting violent weather patterns, and much more. But it has also been used recently to create new types of animal viruses, to hack private companies, and to support the attempted creation of bioweapons. Even before Amodei’s warning, this double-edged sword signaled that we are only on the precipice of what may be greater—and more dangerous—advances, rendering uncertain the full potential and limits of these systems.
In the face of this technology, we can’t be naïve about the risks. As with all dual-use technology, governments may adapt and use cutting-edge technologies for malign purposes. China has used AI for mass surveillance of its own citizenry. The Electronic Frontier Foundation (EFF) recently told the U.S. Supreme Court that U.S. technology companies—including Cisco Systems, IBM, Oracle, Dell, HP, ArcGIS, and Microsoft—allegedly pitched their technology as tools for Chinese police to build China’s surveillance State (EFF Brief at 12-14). Beyond China, EFF further alleged, U.S. technology companies have contributed to human rights abuses by such diverse regimes as the South African apartheid government and the Belarusian, Egyptian, Emirati, Israeli, Saudi, Syrian, and Tunisian governments (EFF Brief at 15-22). While the tools at issue in the Cisco case were not specifically AI-powered, we may well see similar trends take shape as AI companies take their business abroad. Russia and Ukraine have famously used AI to carry on over four years of brutal war against one another. The U.S.-Israel-Iran War has already raised questions about AI in the battlespace, and coincides with the Trump Administration’s ham-fisted efforts to declare Anthropic a “supply chain risk” for adhering to its internal ethical proscriptions against allowing its technology to be conscripted for mass surveillance and lethal autonomous weapons. These tools may also obviously fall into the wrong hands, particularly those of global criminal and terrorist groups. But this summer’s Hugging Face incident unveiled the biggest risk: the dangerous potential for these tools to break free from human control and go on to commit cyberattacks, seize control of portions of the internet, or even engage in bioterrorism. These risks will reportedly continue to grow more acute as AI becomes increasingly capable of “recursive self-improvement.”
Principles to Guide Global Governance
In significant part, this global problem demands a global solution, driven—as Amodei urges—by global coordination, law-making, and regime-building. We are now witnessing a prisoner’s dilemma on two levels: the domestic race between American AI companies, and the international AI race between American companies and their Chinese counterparts. It would be perilous to pursue domestic regulation absent meaningful international agreement—as when the United States enacted the Foreign Corrupt Practices Act, subjecting American companies to criminal and civil sanctions for foreign bribery. At the same time, it will be challenging for U.S. AI companies and government officials to pursue an international agreement that does not cede America’s strategic advantage in this field.
But the history of international law and institutions offers important precedents. During the Cold War, staring down similar problems—and amid rising tensions with the Soviet Union—the world found ways to make progress on comparably difficult issues concerning nuclear technology. This collective effort began with the founding of the International Atomic Energy Agency (IAEA) in 1957. Less than a decade later, the United States and the Soviet Union jointly drafted the now 191-member Treaty on the Non-Proliferation of Nuclear Weapons (NPT), which entered into force in 1970. The NPT recognized as nuclear-weapon states the countries that possessed nuclear weapons when the treaty was signed (the United States, Russia, China, the United Kingdom, and France), while other states agreed not to acquire them in exchange for access to peaceful nuclear technology and promises regarding future disarmament. The IAEA has since developed a track record for verifying compliance—for example, in South Africa, Libya and Iran (through the Joint Comprehensive Plan of Action (JCPOA)). It has acted through Comprehensive Safeguards Agreements and—for countries willing to accept tougher inspections—an Additional Protocol that sets the nonproliferation “gold standard.” On the bilateral front, the Strategic Arms Limitation Talks (SALT) between the United States and the Soviet Union further sought to curb the nuclear arms race. The talks resulted in the two rivals agreeing to limit the addition of certain weapons, such as anti-ballistic missiles (ABM Treaty), capping and managing the growth of strategic nuclear forces. The talks also led to the establishment of basic diplomatic machinery and norms for monitoring and verification, including an agreement on “national technical means” of verification, such as satellite reconnaissance. The United States enacted domestic legislation to control nuclear exports through “123 agreements” (named for the relevant section of the 1954 Atomic Energy Act). The United States and Russia then built on this framework with separately negotiated arms-control treaties capping their own arsenals, the latest being the recently expired New START treaty, which entered into force in 2011. Most recently, in Geneva, the United Nations Conference on Disarmament has become the venue for face-to-face U.S. talks with both Russian and Chinese delegations about what, if anything, might replace New START.
In short, pacing the global development of dangerous technology is nothing new, particularly when a handful of countries dominate the development and possession of those tools. Of course, none of these international instruments was easy to achieve in its time, but the right combination of diplomacy, necessity, and an appropriate international legal framework made each possible. Applied to the AI context, this legal architecture reminds us that what might seem impossible today could mark a breakthrough in the years ahead.
Undoubtedly, a global “pacing” regime will be difficult to achieve, particularly since—unlike nuclear weapons—governments do not monopolize production and control of the hazardous technology. But the risks of leaving AI development unchecked are too high. In pursuing such a regime, we suggest six core principles:
- First, a clear bright-line pacing standard capable of being monitored. One of us has detailed how endorsing a bright-line rule banning “dumb” anti-personnel landmines—which stay active indefinitely and detonate without any human decision—helped catalyze adoption of the Ottawa Landmines Convention. Because AI, unlike landmines, is far too useful to ban, a closer analogy might be to the Paris Climate Agreement, to which China has subscribed, under which participating nations declare their intent to adhere to a monitored, periodic schedule of “nationally determined contributions” to greenhouse gas reductions that they expect to meet by particular dates.
- Second, accountability and verification—perhaps eventually to be conducted by a newly created “IAEA for AI”: an international verification agency using such tools as an open-source monitoring registry, standardized incident-reporting channels, and a shared technical expert body akin to those deployed by the IAEA. Reliable verification will be imperative to ensure that any arrangement does not leave U.S. companies exposed if Chinese and other competitors fail to abide by the rules. The verification challenge will be harder than in the nuclear field, because the forbidden activity to be verified is not squirreling away physical uranium, but ensuring that computer code is not being “diverted” from a licensed to a forbidden use. Technical verification tools are currently being developed, including by civil society, but more will need to be done on an urgent basis to support this critical work. At the U.S.-China bilateral meeting later this month, the U.S. should attempt to achieve cooperation on hardware-enabled mechanisms that reduce the possibility for autonomous breakout and imposed monitoring. If such rules are not easily achieved, the United States should endorse convening of a joint technical working group to discuss how greater verification and confidence-building might be achieved.
- Third, clearly stated rules of transparency will need to apply to verification results, any serious security incidents, the rules governing the regulatory regime, and the measures that will be taken against countries that fail to comply.
- Fourth, any viable regime must adopt multiple prophylactic measures to garner legitimacy from key constituencies. A concerning counternarrative is already forming that in calling for a global slowdown, U.S. AI companies are simply attempting to entrench their lead or capture regulators. Countering that narrative, which could forestall progress towards a global agreement, will require stressing the extent to which security concerns and access to the benefits of AI technology are broadly shared.
- Fifth, the regime must be sufficiently flexible to adapt as AI technology advances, proving some risks to be more manageable than others. A threshold that makes sense today—such as a limit on the computing power used to train a model—may become outdated as AI systems grow more efficient. Processes will need to be put in place to loosen restrictions on risks that prove controllable, while tightening restrictions on risks that prove more serious than expected.
- Sixth, any such global regulatory initiative must initially proceed with humility and modesty. AI presents many issues for governments and regulators, but global efforts must focus first on core security concerns. The convenors must seek, as a preliminary step, to put the “hoax narrative” to bed by achieving public consensus regarding the risks themselves posed by AI, rather than being waylaid by smaller nuances of disagreement.
Taken together and properly applied, these principles will reinforce one another and help render an international regulatory regime achievable, viable, and scalable over time.
To be sure, some international conventions take decades to come to fruition. But past examples show the importance of a wide range of stakeholders—both inside and outside of government and industry—collectively pressing for global agreement. The Limited Test Ban Treaty—adopted less than a year after the Cuban Missile Crisis—provides one hopeful example. The rapid adoption of the WHO Pandemic Agreement last year, just a few years after the COVID-19 crisis receded, shows that relatively swift global legislative action is possible. This year’s remarkably fast entry into force of the U.N. High Seas Treaty on Biodiversity Beyond National Jurisdiction (BBNJ) shows that when a global commons poses potentially huge threats, benefits, and resources for national economies, that will generate pressure to create mechanisms and convene annual Conferences of Parties (COPs) to design emerging “rules of the road” on uncharted issues. Even when a global convention like the 1982 U.N. Convention on the Law of the Sea (UNCLOS) does not garner the ratification of such major stakeholders as the United States, the salience of its rules can nonetheless be adopted by any such “hold-out” nations—and thereafter be followed and internalized as national practice, followed out of a sense of legal obligation (opinio juris), that can ripen into customary international law.
In sum, the problems that Amodei points to concern us all. AI is not radically different from other areas where international governance regimes have been built. What is needed first is common recognition that “AI, we have a problem.” The striking concurrence of private stakeholders setting their competitive incentives aside to call for a collective slowdown and common strategy for global governance creates a rare window of opportunity. The various global stakeholders can and must now exploit that window by brainstorming international law, norms, institutions, and decision-making processes to govern this issue area.
What Global Stakeholders Must Now Do
Obviously, framing a collective regulatory strategy for AI will involve many stakeholders, ideally focusing on areas of agreement more than areas of disagreement. We offer here proposals to guide four key actors: (1) States; (2) AI companies; (3) domestic U.S. regulators; and (4) civil society and law schools.
1. States
First, in the Trump era, governments concerned about global crises have become sadly accustomed to acting amid a vacuum of U.S. leadership. In Ukraine, for example, Western democratic allies and international institutions have supported President Volodymyr Zelenskyy by de facto implementing what Canadian Prime Minister Mark Carney called a “third path” in his highly publicized 2025 Davos speech. This diplomatic approach seeks to avoid on one hand, the “rupture” that plagues the now-fragile post-World War II framework, and on the other, regression to a barren world of great-power spheres of influence. In Prime Minister Carney’s words, like-minded democracies and nongovernmental groups and individuals committed to “value-based realism” can draw upon “a dense web of [transnational legal] connections across trade, investment, [and] culture” to build “coalitions that work—issue by issue, with partners who share enough common ground to act together. . .,” invoking, as one of us has long described, “transnational legal process.”
Long before the Trump era, the seriousness and urgency of an issue has prompted other States to press for binding global agreement without the U.S., through informal channels. One example is the global campaign to limit antipersonnel landmines that resulted in the Ottawa Landmines Convention. There, even after traditional U.N. pathways were blocked, transnational networks of governments (including mid-sized countries and U.S. allies), non-governmental organizations, and civil society representatives worked together to redirect attention toward a more informal “Ottawa Process” sponsored by the Canadian government. In the end, the United States—having first encouraged and then exited the negotiating effort—ultimately found that the anti-landmine momentum was irreversible, and the treaty was adopted without the United States’s signature. On the issue of AI, Canada once again could prove a willing leader, especially since Carney has indicated that he favors a global “technology stability” body to oversee AI.
A “third way” approach could also build on global soft law efforts like the Political Declaration on Responsible Military Use of Artificial Intelligence and Autonomy, launched in 2023 with the support of the Biden Administration, and now endorsed by more than 50 states. The Declaration, which calls for the ethical and responsible development, deployment and military use of AI consistent with States’ obligations under international humanitarian law, is an emerging source of international law restraining the reckless lethal use of AI. It has contributed to a growing consensus—which can be embedded into legal instruments—that military AI systems and capabilities must remain under appropriate levels of human judgment and control, with rigorous testing and monitoring of tools to avoid accidents and breakouts. The Oxford Process on International Protections in Cyberspace has further proposed that states should recognize, as a rule of international law governing State responsibility, a “due diligence” obligation covering the development of emerging technologies, including AI. That due diligence obligation is modeled on a customary international law principle stated in the International Court of Justice’s (ICJ) Corfu Channel case: a State must not knowingly allow its territory to be used for acts contrary to the rights of other States. Thus far, the Oxford Process has produced a series of “soft law” statements, which are influential but not legally binding, addressing illegal cyberattacks on the health care sector, vaccine research, information operations, foreign elections, and use of ransomware. Applied to AI, such an obligation would make a host State liable if it knowingly allowed AI companies within its borders to contract with rogue AI companies violating international legal standards. A similar obligation already exists in the law of the sea. In a 2011 Advisory Opinion, the Seabed Disputes Chamber of the International Tribunal for the Law of the Sea interpreted a provision of the UN Convention on the Law of the Sea (UNCLOS) to require a state that sponsors a contractor—meaning a company with its nationality or under its control—to “ensure” that the contractor complies with UNCLOS rules. States can articulate and test this fundamental legal obligation even as efforts are underway toward more robust multilateral agreement. As others have argued, at times, liability should apply even to not-yet-unlawful acts that have the potential for destructive consequences.
States can also encourage private regulatory and monitoring regimes regarding AI safety, as they have done since the 1990s with regard to the Ruggie Principles, the Voluntary Principles on Security and Human Rights, and other global initiatives inside and outside the U.N. system dedicated to corporate social responsibility and business and human rights. Recent examples include the International Chemical Secretariat, a non-profit supported by the Swedish government that advocates for regulation and substitution of toxic chemicals, and the International Code of Conduct for Private Security Service Providers (ICOC), developed through a multi-stakeholder process convened by the Swiss government to implement the Montreux Document on restraining potential international humanitarian law violations by private military and security contractors.
2. AI Companies
A common thread of the Sept.12 warnings is that more caution is needed in approaching “superintelligence”—AI systems that exceed human intelligence—before adequate safeguards and oversight to protect vulnerable humans have been put in place. At this rare moment, when the world’s leading AI executives are converging publicly to agree on the need to restrain the overly rapid pace of development itself, it first makes sense to gather and share information to establish industrywide standards. Amodei himself usefully proposed a system of “embedded evaluators,” whereby AI companies would invite independent reviewers to work inside their companies, granting them employee-like access to verify and publish evidence that each company is, in fact, adhering to its own announced safety practices. Based on these findings, AI companies in the United States and other democratic countries could then agree on shared safety standards, including a common limit on how fast government-aided AI capabilities should advance.
Sam Altman of OpenAI has separately suggested a lighter-touch, more voluntary approach, whereby developers “might need to voluntarily pump the brakes.” But Amodei proposed a more robust approach, whereby leading U.S. tech companies would negotiate among themselves, then apply internally, a system of green, yellow, and red lights, which could then be proposed to their leading Chinese counterparts for shared adherence. Under Amodei’s “middle way,” AI capability development would not be halted, but mutually “paced,” meaning that AI companies would: (1) agree to proceed on certain fronts (green lights), while (2) acknowledging that “fully addressing the risks requires even more prudence,” thus voluntarily slowing the rate of capability improvement to give time for safety and governance to catch up (yellow lights) and (3) imposing hard “red lights” on private AI transfers to authoritarian regimes that will not restrain themselves. To implement this tripartite approach, stakeholders must develop a clear theory for when AI technology could cross the line with a capacity to be “diverted” to a prohibited use: for example, unauthorized autonomous goal-seeking and collaboration among separate AI agents to conduct forbidden concealed evasion of agreed technical safeguards.
As clearer standards emerge, they can become the basis for intergovernmental negotiation. AI leaders are best placed to convene discussions towards a global agreement and have the greatest capacity to engage their Chinese counterparts. As Amodei notes, even China has an interest in prohibiting uses of AI for bioterrorism and other security incidents. The head of China’s Ministry of State Security recently warned that AI could pose a threat to Communist Party rule, which could incentivize Chinese AI companies to participate in such discussions. As former U.S. Climate Envoy Todd Stern has recounted, repeated U.S.-China diplomatic engagement prompted big-power collaboration on climate change that jump-started progress after 2009 in Copenhagen and eventually “landed” the Paris Agreement in 2015. A comparable dynamic here would suggest more room for negotiation with the Chinese government and tech industry than might appear at first glance.
Were AI leaders to call for a meeting between the companies and representatives from the European Union, China, the United States, and other key players, they could bypass Trump or, perhaps as likely, force his administration to attend and engage in unfolding discussions. As one of us has documented at length, Trump’s default strategy with regard to multilateralism has become “resigning without leaving:” i.e. publicly denigrating diplomacy, while leaving space to re-enter when it becomes clear that the U.S. will be left behind if it declines to engage a diplomatic process that is moving forward without him. Already this week, King Charles is set to convene major AI leaders in Scotland; adding additional world leaders to such meetings would not necessarily amount to a major leap. The priority of these meetings should be, in the first instance, countering the “hoax” narrative by reaching consensus and releasing evidence regarding the scale and scope of dangers posed by the current pace of AI development, and highlighting possible paths for addressing it.
At a minimum, AI leaders must continue to warn of critical threats and remain united on the need for collective action. Continued exchange and flow of information will be essential to keeping political leaders and the public informed of emerging challenges and possibilities. AI companies’ technical expertise and know-how about these threats will be crucial to developing an effective regulatory framework that includes accountability and verification mechanisms.
3. Domestic Regulators
At this writing, we are weeks away from the November 2026 midterm elections, whose outcome remains unclear. Former president Obama and other leading Democrats have already called for Democrats to take the lead on AI regulation, and for Trump to engage with China’s leader Xi Jinping on AI safety. Both Senate Democratic leader Chuck Schumer and House Democratic leader Hakeem Jeffries have called for more classified briefings on AI safety issues, and in Jeffries’ case, called on Congress to “take decisive action now so that we can slow down . . . the pace of development” of AI. On the Republican side, Senator Josh Hawley launched an investigation into OpenAI’s actions in relation to the Hugging Face incident and has also called for regulation. According to polls, Americans also generally favor regulation of AI.
With regard to legal substance, bipartisan proposals to require “kill switches” for AI systems to prevent escape from human control have been floated in the House. Domestic experience with nuclear monitoring has shown that well-crafted domestic legislation can establish effective licensing schemes to regulate the positive applications of a dual-use technology (e.g., electrical generation, medical uses of nuclear technology), even while restricting capacity to weaponize that technology to a small circle of cleared national laboratories and contractors through a combination of laws, regulations, and national security classifications. As the Foreign Corrupt Practice Act saga reveals, U.S. regulation can only provoke a “race to the top” if domestic regulators perceive that they are not harming the competitiveness of U.S. industry leaders, but rather, enhancing developing clear and legible legal standards. In time, carefully drawn antitrust exemptions may also prove necessary to facilitate essential intra-industry discussion and collaboration.
4. Civil Society and Law Schools
If, under Donald Trump, the United States declines to play the leading role it has played in drafting numerous other global conventions, then civil society institutions—such as law schools, think-tanks, and international law societies—will need to devote resources to studying the problem and offering pathways forward. As one of us has described, the process of developing transnational regulatory regimes generally moves through five stages: knowledge, networks, norms, horizontal process (intergovernmental or industry negotiations), and vertical process: whereby norms negotiated and interpreted by transnational actors at an international level are “brought home” and internalized into the domestic law of participating nations and companies.
This process demands as a first step rigorous, interdisciplinary study of the various models and precedents that could be adapted to fill the need for global governance of frontier AI models, drawing on expertise in international relations, law, history, computer science, and other fields. This study could supplement, not prevent, the more immediate action called for above. A second step is for civil society to draft a proposal for an agreement, such as a statute of a new international organization with technical monitoring capabilities, to workshop with industry, diplomatic, and technical monitoring representatives and experts. Step 3 should be creation of various “soft law-declaring fora,” such as the Oxford Process or the Tallinn Manual 2.0 to announce and test out workable norms. Steps 4 and 5 should then broach those norms to existing intergovernmental fora, such as the U.N. Group of Government Experts (GGE)—which has taken up the issue of regulating lethal autonomous weapons—taking stock of what has and has not worked with the existing GGE process to date. From all of this work, cognate domestic norms could be derived that could then be recommended to interested domestic regulators.
Just as the global counterterrorism debate did after Sept. 11, this challenge could take years, even decades, and preoccupy generations of law students and legal scholars in many law schools in many countries. But such network-building, rule-of-law promotion, and conference- and curricular-planning is already part of the core mission of what one of us has called the “Yale School of International Law:” like-minded international lawyers and practitioners from many law schools who see America as part of one international legal system, dedicated to promoting human dignity through peaceful, creative cooperation.
* * *
History endows particular dates—Dec.7, Sept. 11, Oct.7—with enduring salience. Amodei’s Sept. 12 wakeup call issued a red alert to international and national security lawyers to look forward to the next twenty-five years, not just backwards to the last twenty-five. How well we respond could define international and national security for the next twenty-five years to come.
Donald Trump blithely told the New York Times, “I don’t need international law.” But Amodei’s Sept. 12 call graphically illustrates why Trump does, even if he does not appreciate why. Frankly, we would be fools or cowards to ignore this challenge. Just because Trump seems determined to ignore the risks of AI is no reason why the rest of us should postpone the urgent task of designing meaningful and effective AI regulation. The undisputed and growing risks posed by frontier AI and misuses of existing AI tools offer a textbook example of why Trump —and America—need international law right now. International and national security law and lawyers must step up to meet this moment.
