Overview
- Anthropic published a blog post calling for a mechanism to slow or temporarily pause frontier AI development to let alignment research and public oversight catch up.
- The company disclosed that more than 80% of code merged into its production codebase was authored by Claude and that engineers are shipping about eight times more code than in prior years.
- Anthropic says its most advanced model, Mythos, remains in limited preview for defensive testing and has been provided to vetted partners including national cyber teams for vulnerability research.
- The proposal drew mixed reactions with some industry figures praising caution and others, including competitors and commentators, calling the timing suspect as Anthropic pursues large funding and a confidential IPO filing.
- Anthropic warns a useful pause would require agreement and verifiable checks among multiple frontier labs across countries because tracking decentralized compute is far harder than policing physical sites, and regulators and cyber agencies are now engaging with preview access and verification debates.