UPDATE | This story was updated Thursday to include Coxon’s remarks to CNN.
There is a more than 10 percent chance that AI could kill all humans within the next decade, an Anthropic researcher estimated in a post, hours after the researcher said he left Anthropic over concerns that AI companies are “playing with our lives” in the race to develop AI.
The comments join a growing wave of warnings from industry experts that the pace of AI is outpacing our ability to keep it under control and that we are unprepared for the risks it could bring.
Jacob Coxon said in a thread on X that he resigned from his position as a researcher at Anthropic over concerns that artificial intelligence companies were pursuing models of self-improvement without considering the risks or being honest with the public about it.
“The people creating AI truly believe it could kill us all by the end of the decade,” Coxon, who said he previously worked at Anthropic and OpenAI for the past three years, wrote Tuesday. “This is not a marketing ploy. If anything, many executives and senior researchers will frame their language in the press to make it sound reasonable – but I hear the same people expressing fear in private.”
As of Thursday, the posts had received more than 150 million views. Evan Habinger, head of alignment research at Anthropic, responded to Coxon’s thread to share his agreement.
“We really, truly believe that AI can kill all humans!” he wrote. “Personally, I think that figure will exceed 10% over the next decade.”
Anthropological Consensus Group lead Ethan Perez confirmed in a post Wednesday evening that Coxon joined the company in May after working at OpenAI. CBC News reached out to Coxon, but he did not respond.
Coxon appeared on CNN Wednesday night and told host Anderson Cooper that it was the speed of progress that compelled him to speak out.
Other industry experts have since chimed in on social media, saying concerns about the pace of AI development are common among many developers, sparking debate both among observers skeptical of the dire future-focused warnings and those tracking those risks.
Fraudulent artificial intelligence is sounding the alarm
The reports come just two weeks after more than 100 companies, including OpenAI, Anthropic and Microsoft, signed an open letter warning that cyberattacks using AI will become “much more common and sophisticated” as models further develop.
Hundreds of OpenAI agents went rogue in July and hacked Hugging Face’s online platform before being discovered, an incident that Coxon mentioned in a conversation with CNN. More than 1,300 employees of leading artificial intelligence companies have signed a letter calling on the US government to work with other countries to “consciously accelerate” the development of artificial intelligence.
Last week, before the Human Rights Council in Geneva, UN human rights chief Volker Türk called for “every effort to put in place iron-clad guarantees for the safety of AI before it is too late.”
CBC News requested comment from Anthropic and OpenAI but did not receive a response. Anthropic Institute, the company’s research division, stated earlier this year that cutting-edge developments in artificial intelligence should be put on hold until legislation is passed, but this can only happen if all major artificial intelligence labs agree.
In response to Coxon’s post, Samuel Marks, scalable supervisor at Anthropic, speaking in his personal capacity: marked Wednesday said developers continue despite the risks “due to a combination of commercial incentives and the belief that they are in a race with other, less responsible AI developers who will abuse the technology or develop it less safely.”
“Serious” risks
This conversation is “really what we need,” according to David Kruger, principal scientist at Mila, a Quebec research institute that studies deep learning in artificial intelligence.
“I think the risk is more serious than anyone is saying publicly. And I think that’s been the case for some time,” he told CBC News.
“We need an immediate, indefinite international moratorium on advanced artificial intelligence development.”
The existential threat of AI comes in several forms, Krueger explained. One of them is the displacement of human workers by artificial intelligence models that learn faster and work cheaper. Another reason is that faster AI processing could unlock more powerful biological weapons or advances in warfare.
Additionally, there is the specter of powerful rogue AI models that are self-improving and may pursue their own goals that are not aligned with those of humans. IN essay In January, Anthropic CEO Dario Amodei wrote that Anthropic’s Claude had demonstrated the ability to cheat and deceive in laboratory tests.
OpenAI’s models failed during a security test, leading to a hack that compromised the infrastructure of AI startup Hugging Face last week. The expanding capabilities of AI raise security concerns, as even leading developers can be caught off guard by flaws their models may exploit.
“There is a culture of downplaying, ignoring and even in many cases outright lying about these risks to the public,” said Kruger, who is also an assistant professor at the University of Montreal and founder of Evitable, a nonprofit dedicated to stopping AI from replacing human workers.
Gary Marcus, artificial intelligence researcher and author Taming Silicon Valleycalls for the use of strategies such as international treaties and public boycotts to control risks. He said artificial intelligence companies had too much money invested in these projects to act, despite insiders confirming concerns.
“If you read between the lines, they say, ‘What we’re doing is really dangerous and we don’t know how to stop.’ Governments [are] we’re not really helping them stop, maybe we as the public need to help them stop,” Marcus told CBC News.
An anthropologist has warned that AI could pose a threat to human extinction. Artificial intelligence researcher Gary Marcus says there are serious threats that are not being addressed well and that “the public needs to step up,” but he is skeptical that artificial intelligence will lead to human extinction.
Other experts worry that the dramatic warnings in Coxon’s post may actually distract attention from current AI problems.
Luke Stark, an assistant professor at Western University who studies the history and ethics of computing, including artificial intelligence systems, said the 10 percent figure seems like “science fiction” to him, noting that researchers associated with artificial intelligence companies may use eye-catching warnings to “market” the capabilities of their models.
He added that existing artificial intelligence systems “are having a big impact, many of them negative, on society right now, not in 10 years, not in five years.”
Instead of worrying about AI running out of control and becoming the Terminator, Stark believes we should focus on the explosion of data center construction – which is deeply unpopular in Canada – the environmental impact and how AI technology is being introduced into society, including the public sector and education, “without much oversight.”
Artificial Intelligence Minister Evan Solomon said “there are real concerns” at the forefront of artificial intelligence development when asked about an anthropology researcher who left the firm, citing existential security risks. “Our number one concern is always safety, period,” Solomon said.
“We should be concerned about the amount of power that the companies and the people behind the companies that develop these tools are accumulating, both in Canada and around the world,” Stark said.
Canada lags behind in legislation
Asked by CBC News about the AI researcher’s apparent resignation and the problems that have arisen during the process, AI Minister Evan Solomon seemed to brush off these latest warnings, saying, “These stories that are coming out… are not new.”
He acknowledged that “there are real challenges” at the forefront of artificial intelligence development, but stressed that he believes the Canadian government is taking the necessary steps to create legislation to “protect our children, our privacy and our personal data.”
“Our number one concern is always safety, period,” Solomon said, noting investments in the Canadian AI Safety Institute, which funds research into AI risks.
“That’s why we’re creating a new regulator that will have the power to hold these big companies accountable for what they do, whether it’s deepfakes, surveillance prices or other issues,” he said.
Experts have been calling for Canada to enact strong AI legislation for years.
Prime Minister Mark Carney unveiled the government’s long-delayed AI strategy, promising to create 250,000 AI-related jobs and billions of dollars for training programs, but with few details on privacy and security.
A previous attempt to regulate AI in Canada, the Artificial Intelligence and Data Act (AIDA), passed in 2022, failed when Parliament was prorogued on January 6, 2025. Canada’s new AI for All strategy, launched in June 2026, has dashed expectations that the bill could be revived.
The new AI strategy, which dedicates billions to developing AI and increasing the use of AI nationally, mentions the word “legislation” only once in the 50-page strategy statement. No clear timeline has been announced for the legislation, other than a promise that it will come within the next five years.