ANTHROPIC CEO Dario Amodei said in a long post Wednesday that governments should
have the authority to block deployment of new AI models that pose specific or
"unacceptable" risks. He called for mandatory third-party testing to assess
models' risks across multiple domains and compared the current phase of AI
development to cars, planes and drugs — critical to the economy but capable of
causing mass harm if misdesigned or misused. Amodei's remarks are his strongest
to date in favor of tighter AI regulation. Anthropic had warned its MYTHOS model
could find and exploit critical software vulnerabilities and limited access to a
small set of partners; this week the company released an updated version that
removes the model's cyber-attack capabilities.
