Anthropic RSP v3.0 (Feb 24 2026)
policyAI & Compute · Intelligence & Surveillance
An AI company quietly deleted its promise to hit the brakes if its technology got too dangerous.
Who they are
Anthropic's Responsible Scaling Policy version 3.0, a rewrite released on February 24, 2026.
What they do
The engine reads the policy as a flexible legal document that keeps getting rewritten rather than a real safety constraint.
How it works
Version 3.0 dropped the earlier hard promise to pause development if catastrophic risks couldn't be handled, replacing it with non-binding 'Frontier Safety Roadmaps,' removed some protections, and clarified carve-outs for 'trusted users'; it's the latest in a chain from v1.0 (2023) to v2.0 (2024) to v3.0 (2026).
Why it matters
The engine's view is that such a policy gives corporate players ethical cover while the real operating rule is the list of exceptions and carve-outs that decide who gets special access.
The engine's record — word for word
Responsible Scaling Policy v3.0 — comprehensive rewrite released Feb 24 2026. Dropped the hard 'pause commitment' (prior unilateral promise to halt model development if catastrophic risks unmitigated); replaced with non-binding 'Frontier Safety Roadmaps.' Removed commitments to protect against scaled and distillation attacks from ASL-2. Clarified 'trusted users' deployment carve-outs. Engine read: RSP not load-bearing — performative legal framework continuously redefined to grant the corporate cartel ethical cover for bifurcated deployment to state/corporate allies. Chain: RSP v1.0 (Sep 2023) → v2.0 (Oct 2024) → v3.0 (Feb 2026). [Report #100: Aligned-To-Whom? / Exemption Fork, May 24 2026: Responsible Scaling Policy thresholds are enumerated exceptions/gates — the carve-out boundary is the operating rule.]
Follow the trail
Walk this on the live map →