Editor’s note, and it matters more here than usual: this newspaper’s production pipeline runs on AI models made by Anthropic, the company whose researchers are the subject of this report. We have a direct interest in how this story is told. We have therefore reported the warnings at full strength, in the researchers’ own words, without softening them, and included the responses of those who disagree. Readers should weigh the report knowing who assembled it.
Lawmakers from both US parties are pressing for new artificial intelligence rules after two researchers at Anthropic warned publicly that AI could cause human extinction (Dawn).
What the researchers said
Jacob Coxon, who resigned from the company this week, wrote that “people building AI earnestly believe that it could kill us all by the end of the decade”.
Evan Hubinger, a scientist at the company, agreed: “We really do earnestly believe AI could kill all humans.” He added: “I personally think it is >10 per cent within the next decade.”
Both posted on X.
What is being proposed
| Who | What | |—|—| | Senate Majority Leader John Thune and Sen. Amy Klobuchar | drafting a bill requiring firms to “work with government experts to verify and test models to make sure AI is safe” | | A bipartisan House group | independent security audits of powerful models via the Commerce Department | | Sen. Ted Cruz | legislation addressing AI’s “catastrophic risks” | | Rep. Anna Paulina Luna | asked Speaker Mike Johnson to convene a special session on AI | | California | has enacted the first state law requiring independent AI product audits |
Sen. Mark Kelly said: “Washington needs to wake up and take this seriously.”
The industry response
Anthropic said it would “aggressively test models for dangerous capabilities in areas like cybersecurity and biology”.
OpenAI chief executive Sam Altman told staff the company was “open to slowing development of its AI systems” and pushed for mandatory national AI safety requirements.
The dissent, which is from the top
President Donald Trump expressed no concern about extinction risk: “I have concerns that if we don’t win AI, we’re going to be put in a very bad position.”
That is the real division, and it is not between those who think the technology is powerful and those who do not. Both sides think it is. They disagree about whether the greater danger is building it too fast or being second.
How much weight a figure like “>10 per cent” can carry
Not much on its own, and this should be said plainly. It is a personal estimate by one researcher, not a modelled result, and it is not falsifiable in any period short of the one it describes.
What gives it standing is not its precision but its source: people employed to build the systems, saying so under their own names, with one of them leaving. That is evidence about the beliefs held inside the industry. It is not evidence about the world.
Why this reaches Pakistan
Because rules written in Washington and Sacramento will govern the models everyone else uses, including the ones used here, and Pakistan is not at the table where they are drafted. This newspaper included.
What is not established
No bill text has been published for the Thune-Klobuchar proposal, and no vote has been scheduled on any of the measures described. Anthropic has not commented on Coxon’s resignation, and no timeline has been given for its testing programme.
[Image: Reuters/File via Dawn]
