Scientists and protesters greet Anthropic's stock debut by calling it an unsafe AI company
Anthropic spent years selling itself as the careful AI company. As it became a public company this week, a group of prominent scientists, legal scholars and tech founders said that pitch no longer holds. Their statement, organized by the Center for AI Safety (CAIS), a San Francisco nonprofit, puts it bluntly: "Anthropic is not a safe AI company."
For investors and buyers, the issue is the brand, not this week's share price. Anthropic's safety reputation helps it hire researchers, win cautious corporate customers and make its case in Washington. The attack came from an unusually broad coalition. It was paired with a street protest that ended in four arrests at the company's San Francisco headquarters.
The market did not flinch. Shares sold at $83 and closed Thursday at $97.17, up about 17%, according to the San Francisco Standard.
Who signed, and what they claim
The statement went up on Wednesday, November 18, the day before trading began. By Wednesday night, US time, 116 people had signed, according to a count published by SnowRock. The statement's own website now lists more.
The signers include Harvard psychologist Steven Pinker, Nobel economist Oliver Hart, Turing Award winner Judea Pearl, and Harvard law professors Lawrence Lessig and Laurence Tribe. Four Fields Medal mathematicians, including Timothy Gowers and Peter Scholze, are also listed on the statement's website.
What makes the list unusual is who else is on it. Safety campaigners such as MIT's Max Tegmark signed alongside people who usually argue for building AI faster. Those include Guillaume Verdon, founder of the "effective accelerationism" movement, software entrepreneur George Hotz, Hugging Face co-founder Thomas Wolf and xAI co-founder Igor Babuschkin.
The statement makes three main charges. First, it says Anthropic warns that AI able to improve itself, known as recursive self-improvement, could concentrate power and escape human control, yet races to build it and pitches it to investors as an advantage. Second, CAIS says that in February Anthropic weakened its Responsible Scaling Policy, its internal rulebook for when to add safeguards, by dropping a pledge to pause on its own. Third, CAIS says the company lobbied to weaken state AI safety bills in California and New York and privately backed federal rules that would override state laws.
Anthropic described the February rewrite differently at the time. It separated what it would do alone from what it recommends for the whole industry. It said its danger thresholds had proved harder to read than expected, and that government action had been slower than hoped. It has also argued, Dealroom notes, that one company pausing while rivals press ahead could leave the world less safe.
A protest at the front door
On Thursday, about 100 to 120 people rallied outside Anthropic's headquarters, according to Mission Local, the San Francisco Standard and AFP. Their main demand was a ban on self-improving AI. One sign read "Don't invest in extinction," AFP reported.
Four protesters who stayed at the building's entrance were arrested around 2:30 p.m., both Mission Local and the Standard reported. Mission Local said they were blocking the lobby doors. One was Guido Reichstadter, a co-founder of the protest group Stop AI. Police said those arrested would receive a ticket and a court date, then be released.
Speakers included Jacob Coxon, a former Anthropic researcher, and Daniel Kokotajlo, a former OpenAI researcher. Kokotajlo told Mission Local that AI companies could be close to building AI that improves itself: "Probably six months. Maybe it's 12 months." That is his forecast, not an established finding.
At the market, the tone was different. At the opening bell, chief executive Dario Amodei said the company "will not bend to either business or political pressure," according to the Standard.
What Anthropic says, and where the critics are weaker
Anthropic had not publicly answered the statement as of Thursday morning, The Next Web and Dealroom reported. AFP reported that the company says it has not yet reached the stage of AI improving itself.
Its own filings cut both ways. The amended IPO prospectus says Anthropic does not expect to let its Claude models fully build their successors without "meaningful human supervision," according to The Next Web and SnowRock. But SnowRock notes the same document calls AI helping to build AI a structural advantage. Dealroom reports the filing also warned that the company could be targeted over its safety positions.
The coalition has weak spots. Some signers have ties to competitors. Babuschkin co-founded xAI, and CAIS director Dan Hendrycks has advised xAI on safety, Dealroom notes. Wolf said he would not have used the word "reckless" and did not know the statement's title when he signed, though he backed its call for transparency.
Hendrycks also claimed on X that many AI safety groups agreed in private but "felt unable to sign without risking relationships or future funding." He named none of them, and the claim has not been independently verified.
For now the cost to Anthropic is reputational, not financial. But the safety brand is part of what investors bought on Thursday. Three things are worth watching: whether large corporate and government customers start asking about the February policy change, whether lawmakers take up CAIS's lobbying claims, and what Anthropic says in its next public risk report, which its policy now promises every three to six months.
Written by AI from the public sources linked below.