Safety researchers could get in-house access at Anthropic and OpenAI

single source· 1 articles · confidence: low · first seen 2026-09-16 21:07 UTC

What this means for you

Nothing to act on yet: no agreement, no access terms, no named evaluator, no date. This is one outlet's account of an intention, not a policy you can plan around. If you sell evaluation or compliance work to labs, watch whether access terms and independence rules are ever published.

Anthropic and OpenAI want independent safety evaluators placed inside their own labs, according to TechCrunch, which describes it as an intention on both sides rather than a settled arrangement. Researchers the piece draws on welcome the access it would give, but say meaningful oversight also requires the labs to be transparent, the evaluators to be genuinely independent, and eventually regulation. No names, funding arrangement, terms of access or timetable are given.

Key facts

  • ·TechCrunch reports that Anthropic and OpenAI want independent safety evaluators embedded inside their own AI labs. source
  • ·Researchers quoted in the report welcome the access, which they describe as unprecedented. source
  • ·The same researchers say meaningful oversight requires transparency, independence and eventually regulation. source
  • ·The report names no evaluators and gives no funding model, access terms or timetable. source
  • ·The report was published on 16 September 2026. source

What the sources say

  • TechCrunch AI — Reports both labs want outside evaluators placed in-house; researchers welcome the access but doubt it without regulation.

Sources

The original reporting. Follow these — they did the work.

← the wire