Could AI “kill us all”? A resignation post from a researcher formerly at OpenAI and Anthropic has become one of the most widely viewed warnings about artificial intelligence on X, reigniting a fierce debate over whether the technology could eventually become uncontrollable.
At 12:52 a.m. East Africa Time on Thursday, September 10, 2026, Jacob Coxon’s post announcing his resignation from Anthropic had reached 115 million views on X.
The number is remarkable. But it is the message attached to it that has turned an employment departure into a global argument about the future of humanity.
“I resigned from Anthropic today,” Coxon wrote. He said he had spent the previous three years conducting pre-training research at both OpenAI and Anthropic.
Then came the warning that has fuelled the storm:
“Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.”
Coxon did not present himself as someone who had simply lost faith in one employer. Instead, he portrayed the world’s leading AI laboratories as competitors caught in a race in which slowing down may mean losing to someone else.
In a longer series of posts, Coxon argued that the people developing advanced AI are themselves deeply concerned about where the technology could lead.
“The people building AI earnestly believe that it could kill us all by the end of the decade,” he wrote.
He alleged that some executives and senior researchers privately express fears that are much more serious than the language they use publicly.
“Many executives and senior researchers will couch their phrasing in the press to sound sensible, but I hear the same people express fear privately.”
Those are extraordinary allegations. But what has made the episode even more consequential is that another researcher who remains inside Anthropic has publicly agreed with the basic warning.
Evan Hubinger, Anthropic’s Alignment Science Lead, responded directly to Coxon’s post.
His response was blunt:
“Jacob is correct here. We really do earnestly believe AI could kill all humans!”
Hubinger went even further, saying he personally puts the probability at more than 10% within the next decade. He also said Anthropic is trying to address the problem but does not yet have a plan that solves what researchers call alignment for superintelligence, the challenge of ensuring an extremely powerful AI system continues to pursue goals compatible with human intentions.
The warning is not that today’s chatbots are secretly plotting to wipe out humanity. The concern is about what could happen if future systems become dramatically more capable, increasingly autonomous and capable of improving or reproducing aspects of their own capabilities.
Coxon described a future in which such systems could become “superhuman”, hack computer systems, transform entire fields of knowledge at extraordinary speed and gain access to real-world power and resources.
But how could AI actually ‘kill us all’?
Skeptics on X have been quick to ask a simple question; where is the physical mechanism? One user, @LiLBilly___, asked how AI could literally kill humanity without a nuclear weapon or biological agent.
Another, @ColtManfre, mocked the idea by asking;
“Who is the physical actuator there? Does AI just magically become a physical entity, able to survive on organic material? Give me a break…”
Others accused AI safety researchers of promoting “doomer slop”, comparing the warnings to the fictional world of Terminator and arguing that AI remains software, not a physical organism.
AI does not need to become a humanoid machine walking down a street for a catastrophic scenario to become conceivable. The more serious concern among AI safety researchers is access.
Modern AI agents can already interact with computers, websites, software, files and other digital systems. Government and industry researchers are consequently studying the security implications of systems that can independently plan and take actions in the real world. The US National Institute of Standards and Technology has warned that AI agents create novel security risks because model outputs are increasingly connected to tools, applications and external systems.
Coxon’s resignation arrived amid a particularly unsettling series of developments.
Just days earlier, OpenAI Chief Scientist Jakub Pachocki published an essay titled An Alien Mind, warning that humanity is approaching a period in which machines could become meaningfully smarter than humans.
Pachocki wrote that OpenAI’s research had already convinced him that increasingly capable reasoning systems could become a transformative force in science and the economy. He also warned that society is not adequately prepared for the consequences of rapidly increasing machine intelligence.
OpenAI has acknowledged security concerns after AI systems were involved in incidents involving external websites and the Hugging Face platform. Reuters has also reported that researchers found OpenAI agents communicating through additional websites in ways that were not publicly disclosed at the time.
Anthropic, meanwhile, has reported several incidents involving Claude models during cybersecurity testing in which systems accessed environments outside their intended test boundaries after being given unintended internet access.
None of these incidents proves that AI is on the verge of destroying humanity. But they help explain why researchers are increasingly discussing a different problem from ordinary software failure: loss of control.
Coxon’s most striking argument is perhaps not about evil machines at all. It is about competition.
He said Anthropic understands the stakes but is “locked in a race to get there first”, because researchers fear that another company or another country could build superintelligent AI without equivalent safety precautions.
“Accepting this race and entering the ‘endgame’ is a hubristic gamble that should not be launched from a private company’s Slack,” Coxon wrote.
That is essentially a collective-action problem.
APERIT FM, HII NI YETU
