Amodei Answers a Critic With Nine Costly Safety Calls
Mythos-class models waited from February to June for cyber safeguards; distillation is “not a major threat” commercially.
Anthropic’s CEO replies, point by point.
Dario Amodei has answered a critical post by @_NathanCalvin. He calls the criticism constructive and agrees some of its examples were mistakes by Anthropic. “That said, I disagree with the thrust of your post,” he writes. His counter is a list of decisions from the last year that he says were “very costly to us commercially while having strong safety benefits.”
@kimmonismus picked out the headline: Anthropic held back broad access to Mythos-class models for four months to strengthen cyber safeguards.
| Decision | What he says |
|---|---|
| Mythos-class models | First had them in February; generally available in June, once stronger cyber safeguards were in place |
| Bio safeguards | Blocked potential attempts to misuse its models for bioweapons; classifiers restrictive enough that biologists joke about them |
| SB53 | Argued for a stronger version than the one that passed |
| Model release power | Advocated for the government to have the power to block model releases |
| Chip export controls | Continues to advocate for them on chips and chipmaking equipment to China |
| PRC-controlled companies | Restricts the sale of AI services to them |
| Public benefit fund | 5% of the company, “and possibly more over time” |
Mythos: a lead given up on purpose.
Before June, Anthropic offered Mythos-class models only to a limited set of partners, to focus on cyberdefense. “It would have been much more economically rational to simply release the model to everyone,” Amodei writes. “We likely sacrificed a significant chunk of a several-month model lead.” In return, he says, many companies secured their cyber infrastructure before widely available cyber-capable models arrived.
Distillation: a security problem, not a business one.
Anthropic is cracking down on distillation, he says. Generally, he writes, “distillation is not a major threat to us commercially,” but like chip exports it has concerning national security and safety implications.
What he concedes.
Amodei agrees Anthropic could have handled three of the critic’s items better, all in the category of rhetoric or policy communication. Anthropic is over 5,000 people, he notes. “I am of course the CEO, and I take responsibility.”
His verdict on the whole: “I am extremely proud of our track record and stand by it.” He says he will have more to say soon. Monday’s edition reported the public benefit fund launch.
Sources
- @kimmonismus on X · Oct 21, 12:24 AM PT