Governance and control

US AI Safety Institute secures predeployment model access

NIST signed agreements with Anthropic and OpenAI establishing formal safety research and testing, including access to major new models before and after public release and feedback on potential safety improvements.

0 comments · 0 votesOpen discussion

Public discussion is readable by everyone. Sign in to comment, reply, or vote.

No comments yet. Start the discussion.

CURRENT ASSESSMENT · REVISION 1
AWAY FROM DOOM52confidence 99/100

Why it moved the index

Predeployment government access creates an institutional channel for independent frontier-model evaluation and safety feedback before public exposure, materially strengthening oversight despite the agreements' collaborative rather than coercive form.

AUDIT TRAIL

Assessment history

  1. R1
    Away 52 · confidence 99

    New August 2024 frontier-model governance and predeployment testing agreement.

    12 Aug 2026
SHARE THE FINDINGS

Share this page

DoomBench social sharing card for US AI Safety Institute secures predeployment model access.
  1. DoomBench assesses “US AI Safety Institute secures predeployment model access” as evidence moving away from doom, with magnitude 52 and confidence 99 out of 100 in the governance and control category.

  2. The DoomBench assessment of “US AI Safety Institute secures predeployment model access” is based on reporting from National Institute of Standards and Technology and records the editorial rationale, source quality, attribution, and...

  3. DoomBench summarizes “US AI Safety Institute secures predeployment model access” as follows: NIST signed agreements with Anthropic and OpenAI establishing formal safety research and testing, including access to major new models before...

    https://www.doombench.com/news/us-ai-safety-institute-secures-predeployment-model-access-2024-08-29