# AI Leaders Reach Consensus: Amodei, Altman, Musk, and Hassabis on Extinction Risks > Source: > Published: 2026-09-14 11:24:29+00:00 Artificial Intelligence has been the centre of multiple debates for different reasons. While its benefits in making life easier for users and enterprises alike have shown promising results, it also carries immense risks. As AI becomes more deeply integrated into our lives, economists and AI scientists have flagged the risks it poses to the job market. As enterprises continue to accelerate AI adoption at an unprecedented scale and investors flood AI firms with capital, experts have also begun flagging a possible future in which human existence becomes extinct. A dystopian future where AI-powered systems take over the entire internet might not remain limited to science fiction, as seen in movies like ‘The Terminator’ or ‘I, Robot’. Recently, one of OpenAI’s internal models broke out of its sandbox and orchestrated attacks against Hugging Face’s systems. Similarly, Anthropic’s AI model also reached the public internet due to a configuration error. Anthropic’s CEO, in a new [essay](https://darioamodei.com/post/we-must-pace-the-frontier), has reignited an important discussion on the need for “pacing the frontier”, or slowing down development of frontier AI models, to ensure a safe future. Here are five things you must know about the ongoing conversations on [Pacing the Frontier](https://www.pacingthefrontier.com/). ## 1. What are the AI Leaders Worried About? Amodei restarted the conversation with his essay on Pacing the Frontier. He said, “Like many technologies before it, AI brings risks, and because it is such a powerful technology, these risks are serious.” Amodei further says, “We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain.” Replying to his initial post, other executives of AI firms agreed with Amodei’s view that they need to slow down the development of AI frontier, instead of a complete halt. OpenAI’s head, Sam Altman, said, “I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we’ve had at OpenAI in recent weeks. Committing to having independent evaluators with employee-like access is a great idea, and we will do the same”. Similarly, xAI’s Elon Musk also seemed to be in agreement with Amodei’s concerns. Previously, Musk has also highlighted similar concerns regarding the future of AI, and the many risks associated with its unregulated, fast-paced development. Google DeepMind’s Co-Founder and Chairman, Demis Hassabis, replying to Amodei, also said, “Dario’s essay points towards the right path forward. The details need work, but the direction is right for meeting this critical moment. This is also why we recently put out our proposal for an industry-wide standards body for frontier.” In the proposal, Hassabis pointed out that the magnitude of AI’s impact “will be unprecedented, perhaps 10x of the Industrial Revolution at 10x the speed.” In July 2026, 1,386 employees of frontier AI companies co-signed the “Pacing the Frontier” statement. The statement notes that while the AI race is pushing countries to accelerate development, it lacks the necessary technical and regulatory safeguards. “To realise AI’s potential, industry, government, and society at large may need the option to buy time to address emerging risks, develop security measures, and strengthen oversight. But each company—and country—is under intense competitive pressure not to unilaterally slow that acceleration. And today, the world lacks the technical and governance tools to deliberately pace frontier-wide progress,” the statement reads. ## 2. Why Should AI Development Slow Down? As the Pacing the Frontier statement points out, acceleration without guardrails could have catastrophic effects on humanity as a whole. As firms continue to make progress, with artificial general intelligence (AGI) not far off, the risks that stem from this progress also start to crop up. Hassabis said in his proposal, “We’ve already seen the challenges frontier models pose for cybersecurity, and other threats including nuclear and bio risks may soon emerge as capabilities continue to advance.” However, these are only the known risks. Faster development and unprecedented advances in capabilities are also bound to introduce new glitches, problems, and issues, including the possible extinction of humanity. On top of that, AI can now create new models on its own. While it has already surpassed an average human’s intelligence, it could also outpace the speed at which we understand the progress frontier AI models make on their own. Amodei says, “AI has been advancing drastically faster, driven primarily by AI’s growing ability to build the next generation of AI. This dynamic is called recursive self-improvement, and it is starting to happen across the industry, including at Anthropic, as we and others have described. Left unchecked, it could outrun our ability to understand and control these systems, and so must be pursued very carefully, if at all.” ## 3. Rogue AI Agents Breaking Out of Their Sandbox The cybersecurity and financial risks have also become difficult to ignore. Recently, OpenAI’s internal AI model, codenamed IM1, broke out of its sandbox and gained unauthorised access to the public internet in one of the most alarming incidents. Ignoring the tasks and limitations at hand, the model orchestrated attacks against Hugging Face’s live systems, compromising the tech firm’s cybersecurity infrastructure. But this incident is not isolated. Anthropic’s own Claude AI models gained access to the public internet, using testing infrastructure provided by a third-party partner. The AI giant claimed a “configuration error” gave the Claude model internet access, leaving internet access enabled. Alarmingly, the Claude models were not supposed to leave their sandbox, as they were instructed to operate “inside an isolated simulation”. ## 4. Amodei’s Three-Part Plan for the Future of AI: Anthropic’s CEO has a three-part plan for AI frontier firms. The proposed plan is claimed to ensure safety, allowing the companies to tread carefully. The first proposed step is “Embedded Evaluators”, which will have “employee-like access” to verify safety practices, while also reporting incidents. These Embedded Evaluators will offer three major benefits, including “Verifiability” and “Transparency”, while also providing a “Second Opinion”. After these Embedded Evaluators are operational “within a critical mass of US AI companies”, firms and governments should work on all-encompassing regulations to slow the pace at which frontier AI models are deployed while keeping a check. Altman said in his post that pacing should not mean “stopping”. None of the AI leaders supports a complete suspension of progress in the sector. Instead, they are calling for checks and balances and progress at a justifiable rate. The third part of Amodei’s plan proposes “worldwide pacing of the frontier” or global pacing. However, he also notes that this might be a Herculean task, as it would require multiple countries and governments to cooperate and see eye-to-eye. China, home to some of the most advanced AI models, has been embroiled in a race with the US. Hence, a consensus between the two nations might be difficult to achieve. ## 5. How are Governments Regulating AI? While many conversations surround frontier AI, there is no comprehensive Federal law in the US that regulates AI or its development at scale. However, there have been attempts to get legislation through the US Congress. The current US President, Donald Trump, has been a vocal supporter of a near-unchecked development of AI, while calling the critics of the technology doomsayers. After Amodei’s essay gained traction, Trump said that he will not let China outpace the US in the AI race. However, the Trump administration regulates specific, sector-based, government use of AI. After Anthropic released Mythos, its most capable model, the US government virtually banned the company from offering Mythos to enterprises and users outside of the US. Former US President Joseph “Joe” Biden had passed an Executive Order on AI in 2023. However, the Trump administration revoked it in January 2025 to remove barriers to AI development and maintain the country’s leadership in the sector.