Dario Amodei, CEO of Anthropic, published a landmark essay called ‘Policy on the AI Exponential.’ It is the most serious, detailed, and urgent call to action any frontier AI CEO has ever made to governments worldwide. He is not asking politely. He is warning that the window to act is closing. Here is what he said, why it matters, and what it means for the UAE.
Who Dario Amodei Is and Why This Essay Matters
Dario Amodei is the CEO and co-founder of Anthropic, the company that makes Claude. Before Anthropic, he was Vice President of Research at OpenAI. He has a PhD in computational neuroscience from Princeton. He is one of the few people on earth who has spent the last decade at the absolute frontier of AI development and has thought seriously about what it could do to the world.
He does not write often. When he does, people in AI, policy, and finance pay close attention. His 2024 essay ‘Machines of Loving Grace’ predicted AI could compress a century of biological progress into five to ten years, cure most cancers, prevent Alzheimer’s, and potentially double human lifespan. It was dismissed by some as utopian. By most measures, the trajectory he described has arrived on schedule.
Today he published something different. Not a vision of the future. A warning about the present.
| In only four years, AI models have gone from barely being able to write a coherent line of code to writing most of the code at major AI companies. If scaling laws continue for only a year or two longer, we are likely to get what I have called Powerful AI, or a country of geniuses in a datacenter. Dario Amodei, June 10, 2026. |
The Treebeard Problem: Why This Is Urgent Right Now
Amodei opens the essay with a metaphor that is worth understanding. In The Lord of the Rings, two Hobbits try to convince Treebeard, a slow-moving sentient tree, to defend his forest from an army that is already cutting it down. The problem is speed. The threat is moving faster than Treebeard can respond.
That, Amodei says, is exactly where governments are right now with AI. AI is moving at the speed of a fast-moving army. Governments are moving at the speed of Treebeard. In the several years it can take a legislature to pass a law, AI can go from an amusing toy to what he calls the full country of geniuses.
He acknowledges this was not a failure of governments. Until recently, the risks were real but still theoretical. It was hard to convince policymakers that anything other than a hands-off approach made sense when what they could see was an impressive but relatively mundane technology. That argument is now gone.
The specific event that changed everything: the release of Claude Mythos Preview. Amodei says Mythos Preview scrambled the global cybersecurity landscape and proved beyond doubt that AI models are now tools of global and national strategic consequence. He reveals that frontier models now pose very real risks to cybersecurity and could disrupt the financial sector, critical infrastructure, and national security. He believes biological risks will follow, and serious AI autonomy risks may not be far behind.
| This is the CEO of Anthropic saying, in public, that the AI his own company is building is now a strategic threat to global security. That is not a competitor making accusations. That is the builder issuing a warning. |
The Five Policy Areas He Wants Governments to Act On Now
The essay covers five specific areas. Amodei is not vague. He is specific, detailed, and in several cases has attached actual legislative proposals and financial backing.
1. Regulation and Public Safety
Amodei argues that AI has crossed the threshold where transparency rules are no longer enough. The right model now, he says, is the FAA for aviation. Before a plane can fly commercially, it must pass mandatory technical testing. Before a frontier AI model is deployed, it should too.
His specific proposals: any AI model trained above a certain compute threshold must undergo mandatory third-party testing in four specific areas. Cybersecurity risk. Biological weapons risk. Loss of control of AI systems. Automated AI research and development that could accelerate the other three risks. Governments should have the power to block deployment if a model fails.
He is careful to add safeguards. The blocking power must be narrowly scoped to those four specific risks. There must be protections against political favouritism. He does not want governments using safety powers to pick winners in the AI industry. He wants them used for genuine safety threats only.
Alongside the essay, Anthropic released a legislative proposal on frontier model testing. This is not a think piece. It is a draft law.
2. Macroeconomics and Job Displacement
Amodei has been warning about AI and jobs in public for months. He is careful to separate two things. First, AI will probably cause significant job displacement and we should not pretend otherwise. Second, we should do everything possible to slow and reduce that displacement, not accelerate it.
He is direct about why this matters economically. If AI can do most cognitive tasks better than humans, economic growth could become extraordinarily rapid. But the same technology could make the AI-driven job displacement more severe and more permanent than anything previous technologies have caused. The risk is ending up in a world where economic growth is enormous but almost entirely captured by capital rather than labour.
His policy proposals are specific. Measurement and tracking of AI job displacement through expanded government economic statistics. Wage insurance to compensate workers who have to take lower-paying jobs. Retention tax incentives for employers who avoid layoffs. Workforce training grants. Long-term income support and universal basic income if displacement proves permanent, financed through taxes on AI companies and capital gains.
Alongside the essay, Anthropic also released a policy framework for job displacement, with financial backing committed. This is Anthropic putting real money behind the warning, not just words.
| There is a decent possibility that, despite all our efforts, AI still causes significant enduring job loss. This may be an intrinsic property of the technology and the way it broadly replicates human cognition. Dario Amodei. |
3. Accelerating the Good Stuff: Biomedical Innovation
While sections one and two are about managing AI’s risks, section three is about making sure AI’s benefits arrive fast enough. Amodei argues that the regulatory systems for biomedical innovation were designed for a slower world. AI is about to flood the FDA and EMA with drug candidates at a pace those agencies are not built to handle.
His argument is specific. AI will greatly increase the number of new drug candidates entering the regulatory pipeline. It will improve their safety profiles and effect sizes. It will create treatments for diseases that have never had effective therapies before. It may create entirely new categories of treatment in the way that antibiotics and cell therapies have.
If the regulatory system is not reformed in parallel, the benefits arrive slowly while the risks land fast. He proposes that regulators develop standards now for AI-based drug modelling, AI-predicted toxicology, synthetic control arms in clinical trials, and surrogate endpoints in aging and neurodegeneration research. The goal is not to reduce safety standards. It is to update the tools used to establish them.
4. Civil Liberties and the Risk of AI-Powered Tyranny
This is the section that will generate the most debate and the most attention. Amodei argues plainly that powerful AI in the wrong hands could be the ultimate tool of autocracy. He is not talking about China abstractly. He is talking about any government, including democracies.
His specific concerns: a fully automated drone army that could obey unlawful orders without the conscience check that human soldiers provide. A surveillance AI that could analyze publicly available information at massive scale and infer the innermost details of every citizen’s life without triggering existing civil liberties laws, because those laws were written before this technology existed.
His proposals are equally specific. Autonomous weapons systems should be required to respond to constitutional and judicial oversight, not blindly follow commands. They should be banned entirely for domestic law enforcement use. The data broker loophole that allows the government to purchase private data and use it for mass surveillance should be closed. Any person facing adverse government action should have the right to access AI tools at least as capable as whatever the government is using against them.
And then he says something remarkable. He says that governments are not the only threat. Companies including, explicitly, his own company, could become powerful enough to capture the state or adopt quasi-state characteristics. He says AI cannot safely be fully entrusted to either governments or companies. Both need checks and balances.
| The CEO of an AI company saying publicly that his own company should not be fully trusted with AI, and that independent governance mechanisms are necessary, is not something the industry has said before. It deserves to be read carefully. |
5. A Global Coalition of Democracies
The final section is geopolitical. Amodei argues that AI is not a trade policy issue. It is not a question of how to spread technology around the world. It is a question of who controls the most powerful force in human history and under what values.
His analogy is stark. A nation with powerful AI facing one without it, or even one that is behind by three years, could be the equivalent of an army of World War II Marines facing an army of medieval swordsmen. He argues democracies need to form a formal coalition around AI, built on shared values, that shares chips and manufacturing equipment among members, coordinates safety standards, coordinates economic support, and denies the critical components of AI development to adversarial states.
He wants export controls on frontier chips and semiconductor manufacturing equipment expanded and coordinated internationally. He wants members to commit to rejecting AI-powered domestic repression as a condition of coalition membership. He wants the coalition to start with ideologically aligned democracies and grow iteratively, making membership more and more attractive and non-membership more and more costly.
What This Means for the UAE Specifically
The UAE is not a peripheral player in the world Amodei is describing. It is near the centre of it.
Anthropic’s own Claude models, including Mythos Preview, are being used by UAE organizations. G42, Mubadala, Aldar, MBZUAI, and Khalifa University are all deploying OpenAI and Anthropic technologies under the UAE Vision 2031 strategy. OpenAI is building a major computing facility in the UAE as part of its OpenAI for Countries initiative. Anthropic has navigated complex questions about UAE investment given the country’s governance structure, as reported by The Intercept in June 2026.
The coalition question in section five is the most directly relevant to the UAE. Amodei’s framework calls for a coalition of democracies that shares AI capability among members and denies it to adversaries. The UAE is neither a liberal democracy in the Western sense nor a geopolitical adversary of the United States. It is something in between, and the question of where the UAE sits in a world organised around AI coalitions is one of the most consequential foreign policy questions the country will face in the next three years.
The civil liberties section is also directly relevant. Amodei explicitly warns against AI-powered surveillance and the use of AI to entrench government power domestically. He calls for these to be prerequisites for coalition membership. As a country with significant existing digital surveillance infrastructure and a non-democratic governance model, the UAE will need to navigate these questions carefully as it pursues its AI strategy.
On jobs, the UAE context is specific and important. Mohamed Alabbar announced earlier this year that Noon would cut its workforce by 50 percent in three months due to AI. Amodei’s essay names this exact dynamic and argues governments must begin tracking it immediately and prepare income support mechanisms before the displacement becomes unmanageable. The UAE government has not yet released a specific policy framework for AI-driven labour displacement, and Amodei’s essay is a direct signal that this gap needs to close.
| Amodei Policy Area | UAE Relevance |
| AI Safety Regulation | UAE does not yet have a mandatory pre-deployment testing framework for frontier AI models. TDRA and CBUAE have sandbox frameworks but not binding safety rules at Amodei’s proposed level. |
| Job Displacement | Noon announced 50% workforce reduction due to AI in 2026. UAE has no published job displacement tracking or income support framework for AI-affected workers. |
| Biomedical Acceleration | DHA and DOH are deploying AI diagnostics. The UAE has an opportunity to be among the first to reform its regulatory approval framework for AI-driven drug development. |
| Civil Liberties | UAE has extensive digital surveillance infrastructure. Amodei’s proposed coalition requirements include commitments against AI-powered domestic repression as a membership condition. |
| AI Coalition Geopolitics | UAE sits between Western democracies and China in its relationships. How it navigates the coalition framework Amodei proposes will define its AI strategy for the next decade. |
The Window Is Open. The Question Is Whether Governments Walk Through It.
Amodei ends the essay on a note that is cautiously optimistic. He says policymakers are showing unusual openness to action. He says the combination of clear AI risks, early economic disruption, and genuine public concern has created a moment where forward-looking policy can be adopted faster than usual. He says Treebeard and his forest are waking up.
But he is also clear that being a year out of step with AI progress is not a small gap. AI does not wait for legislative calendars. It does not pause for committee hearings. The essay is not a reassurance. It is a timeline.
For UAE residents, the essay matters for a concrete reason. The country is not a passive observer of the AI transformation Amodei is describing. It is one of the most active participants. It has the highest AI adoption rate in the world. It is hosting OpenAI’s infrastructure. Its companies are cutting workforces in the name of AI efficiency. Its government has committed to AI-powered public services. What happens in the policy frameworks that Amodei is calling for will directly shape the conditions under which residents work, earn, spend, and live.
| The essay is published. The legislative proposals are attached. The financial backing is committed. The question now is whether governments move faster than they have ever moved for a technology issue before. Amodei has said clearly what happens if they do not. |
Sources: Dario Amodei, ‘Policy on the AI Exponential,’ published June 10, 2026 (darioamodei.com). Anthropic legislative proposal on frontier model testing (June 10, 2026). Anthropic policy framework for job displacement (June 10, 2026). Dario Amodei, ‘Machines of Loving Grace,’ October 2024. The Intercept, Anthropic and Gulf State investment reporting (June 6, 2026). Khaleej Times, UAE AI adoption and OpenAI data residency (November 2025). Microsoft AI Economy Institute Q1 2026.
Robius.news — Dubai, UAE — 2026 | Built to be first. Built to be trusted.
