“AI could kill us all by the end of the decade”: Former Anthropic researcher’s warning sparks fresh debate over how fast the technology is moving
What if the biggest danger from artificial intelligence is not what today’s systems can do, but what future systems might be capable of doing on their own?
That question has gained fresh attention after AI researcher Jacob Coxon resigned from Anthropic, warning that people developing advanced AI are seriously concerned about where the technology could lead.
Coxon’s statement has added to a growing debate within the AI industry over whether the race to build increasingly powerful systems is moving faster than research into their safety.
His concern is focused on a future in which AI becomes capable of handling increasingly complex technical work, improving its own capabilities and potentially contributing to the development of more advanced AI systems.
What Is Jacob Coxon Worried About?
Coxon is not suggesting that the AI tools people use every day are already capable of taking control of the world.
His concern is about what could happen if future systems become dramatically more capable.
He has pointed to areas such as coding, mathematics, hacking and scientific research, where increasingly advanced AI could potentially perform tasks that currently require highly skilled humans.
The more important question, however, is what happens if AI becomes capable of helping humans build better versions of AI itself.
That could accelerate development and make it harder for researchers to understand exactly what increasingly advanced systems are doing.
For Coxon, that possibility is one reason AI safety needs to be considered alongside AI development rather than after it.
Why the AI Race Matters
The debate is also being shaped by competition.
AI companies are racing to develop more capable models, while governments see advanced AI as strategically important for their economies, cybersecurity and national security.
That creates pressure to move quickly.
Coxon’s concern is that safety research may not always move at the same speed as capability development. If future systems become more autonomous before researchers understand how to reliably monitor and control them, the risks could become harder to manage.
This remains a future scenario, not an established outcome.
But recent developments show why the question is receiving more attention.
AI Is Already Helping Build AI
AI is increasingly becoming part of the research process itself.
Anthropic has said its Claude systems were involved in around 90% of the company’s research and development work alongside human researchers, while Claude was leading roughly 26% of its R&D projects as of August. The company said human researchers continued to supervise the work.
The development does not mean AI is independently creating the next generation of AI.
But it shows how quickly the role of these systems is changing. AI can now assist with coding, research, analysis and other technical tasks that were previously carried out almost entirely by people.
That makes questions about oversight increasingly important.
What Does the Latest AI Safety Research Say?
The International AI Safety Report 2026 provides useful context to the debate.
The report says current AI systems show early signs of capabilities relevant to potential loss-of-control scenarios, but they are not currently at the level required for such an outcome. It identifies capabilities such as evading oversight, carrying out long-term plans and preventing countermeasures as important factors in assessing future risks.
The report also notes that uncertainty remains significant because the likelihood of these scenarios depends on how AI capabilities and deployment develop in the future.
That distinction is important when considering Coxon’s warning.
His statement describes a serious possible risk. The scientific literature does not establish that the outcome he describes is inevitable or that it will happen by a particular date.
Where Does the United States Stand?
The U.S. government is pursuing rapid AI development while also addressing cybersecurity and national-security risks.
A June 2026 White House executive order called for advancing AI innovation and security, including stronger cybersecurity and protection of critical infrastructure. It also stated that the administration does not intend to create mandatory federal licensing or pre-clearance requirements for AI models.
The White House’s March 2026 national AI framework similarly emphasised innovation, economic competitiveness and American leadership in AI.
This approach places considerable emphasis on maintaining technological leadership while addressing specific security risks.
That is an important part of the wider debate because Coxon’s warning focuses more heavily on the possibility that increasingly capable AI could eventually become difficult to control.
What Is the UN Saying?
The United Nations is approaching AI from an international governance perspective.
In July 2026, the UN held its first Global Dialogue on AI Governance in Geneva, bringing governments and other stakeholders together to discuss international cooperation and AI governance. The dialogue included discussions around safe and trustworthy AI, transparency, accountability and human oversight.
The UN has also established an Independent International Scientific Panel on AI to strengthen the scientific basis for international discussions about the technology.
At the opening of the Geneva dialogue, UN Secretary-General António Guterres said AI was advancing rapidly and argued that countries need to work together on how the technology is governed.
The UN approach therefore looks beyond individual companies or countries, focusing on international cooperation and common governance principles.
The Question Behind Coxon’s Warning
Coxon’s resignation has brought a difficult question back into the centre of the AI debate:
Are we developing increasingly capable AI faster than we are learning how to manage its risks?
There is no simple answer.
AI is already being used for research, software development, business, education and scientific work. At the same time, researchers are studying what could happen if future systems become substantially more autonomous.
The 2026 International AI Safety Report does not say that loss of control is inevitable. The U.S. is continuing to prioritise AI development and technological leadership, while the UN is working toward international cooperation and governance.
Coxon’s warning sits within that much larger debate.
The question now is not simply how intelligent AI can become.
It is whether human understanding, oversight and safeguards can keep pace with the technology as it becomes more capable.
