← Back to live feed1 story
Anthropic's August 14 risk report identifies an internal system called "Model 2" that is more capable than its Mythos 5 model but will not be released to the public. The company raised its misalignment risk rating from "very low" to "low" citing uncertainty following recent cyber evaluation incidents.
The stronger internal model provides a "noticeable improvement" for coding, data generation and agentic tasks. This decision to restrict access aligns with recurring statements from company leadership that certain frontier systems are too dangerous to release to the public.
Sign in to suggest edits
Key sources
- SOURCE@intcyberdigest“"noticeable improvement" for internal work: coding, data generation and agentic tasks”x.com
- SUPPORT@jun_song“How many times have we heard this exact same playbook since he was at OpenAI?”x.com
- SUPPORT@axios“Anthropic sees AI risks rising”x.com
- SUPPORT@gerritd“The company accidentally let 50,000 contractors access its models without any biorisk guardrails for 11 months”x.com
- SUPPORT@gerritd“Anthropic has been very tight-lipped about the Irregular incidents, much more so than OpenAI has been about its Hugging Face hack”x.com
- SUPPORT@gerritd“Anthropic says it hasn't been able to review the transcripts from AISI yet”x.com