Anthropic released an updated AI model safety framework this week that it says no longer commits the company to unilaterally pausing development to reduce catastrophic risk if competitors move ahead without comparable safeguards as the company cites what it describes as a growing "collective action problem" in frontier AI development.
In Version 3.0 of its Responsible Scaling Policy, the company says that "the overall level of catastrophic risk from AI depends on the actions of multiple AI developers, not just one," and that from a societal perspective "what matters is the risk to the ecosystem as a whole."
The policy defines catastrophic risk in plain terms as "risks of the most severe potential harms from advanced AI, such as existential threats or fundamental destabilization of global systems."
