Responsible Scaling Policy v3.0
RSP v3 splits what Anthropic will do unilaterally from what it says the industry needs, and adds a public Frontier Safety Roadmap and Risk Reports.
Anthropic says pre-set capability thresholds proved ambiguous (especially biology), the political climate turned anti-regulatory, and high-ASL safeguards may be impossible unilaterally. The revision separates company commitments from industry recommendations, sets 'ambitious but achievable' public roadmap goals it will grade itself on, and adds Risk Reports with external review.
- Date
- Tuesday, 24 February 2026
- Lab
- Anthropic
- Kind
- paper
- Access
- paper only
Third version after v1 (2023-09-19) and the 2024-10-15 update. Anthropic's rationale (ambiguity of thresholds, anti-regulatory climate, RAND SL5 'not currently possible') is its own account; critics' views were not reviewed. Only the first two-thirds of the post was read in full.
Sources
This record was checked against its sources on 6 October 2026. How we check