Anthropic's RSP bars external evaluators with close personal ties to anyone at the company
morqon · x · 2026-10-22
A leaked/examined section (3.6.1) of Anthropic's Responsible Scaling Policy spells out how external reviewers are selected: they must have significant expertise in evaluations of dangerous AI capabilities, including how such evaluations can mislead (e.g., alignment faking); they must have reputational incentives to be candid about risks rather than depend on Anthropic's continued business; and they must have no conflicts of interest — neither the reviewing organization nor anyone in the reviewers' reporting chain may hold a financial interest in Anthropic or have close personal relationships (family, romantic, or shared living arrangements) with anyone at the company. External reviewer selection is consulted with the Board and approved by the LTBT.
More from Safety
- Upstream CEO: hacked robots can turn dangerous without losing control — TechNadu · 2026-10-22
- Cloudflare's AI security-audit skill tops GitHub with 38.5k stars — alex_verem · 2026-10-22
- Major US banks detect unauthorized third-party AI agents inside customer accounts, Axios reports — EthanJPerez · 2026-10-22
- Seth Lazar: AI Agents Are Already Straining Democratic Resilience — sethlazar · 2026-10-22
- Ben Bajarin: cybersecurity becomes the budget that lets enterprise AI scale — BenBajarin · 2026-10-22
- Europe cannot rent its way to AI sovereignty, Air Street Capital argues — nathanbenaich · 2026-10-22