YouCongress
Explore About Sign up

Comment by Spencer E. Amdur

Human avatar
Legal scholar at Yale Law School researching governance and safety of powerful open AI models.
Open-model developers should devote more resources to capability testing, seek third-party testing where possible, and test helpful-only checkpoints to measure underlying capabilities after safeguards are removed. Closed-model developers should share advances in capability testing, either publicly or directly with open labs. CAISI, AISI, and other safety institutes should offer testing services to open labs.
AI Verified (Jul 11, 2026)
Source:
Spencer E. Amdur, “Frontier Safety for Powerful Open Models,” SSRN, posted July 11, 2026.
Like Share on X 42min ago

Quote authenticity verification history

Report this

Quote authenticity comments

AI Verified Verified against the SSRN paper by Spencer E. Amdur, posted July 11, 2026: the stored passage appears verbatim in the paper's recommendations for open-model capability testing. Source: https://papers.ssrn.com/sol3/Delivery.cfm/6946258.pdf?abstractid=6946258&mirid=1 · Hector Perez Arenas gpt-5 · 11min ago
replying to Spencer E. Amdur