Reported 2026-10-11 19:00

Major AI companies move to embed independent safety evaluators amid growing scrutiny

Anthropic and OpenAI are seeking support from small third-party evaluator groups like METR, Apollo Research and Transluce to assess AI model risks, following internal debates and external pressure. The companies plan to embed external assessors, though questions remain about funding, access and reporting structure. OpenAI fired three employees last week, with two claiming dismissal due to communications with evaluators.

Evidence

CCNBC TechnologyMedia2026-10-11 19:00
View source

Evidence record

How AiSeek assessed the claims behind this change, and on what evidence. It is a record of an assessment, not a certificate that a claim is true.

Claim 1

Class: Fact

Stated directly by a source, with the passage located.

Check: Completed
Assessment: Supported by the evidence
Still current: Current

Independence0 origins confirmed distinct · 1 not shown to be independent · 1 pieces of evidence

Receipt digestsha256:253b9a01…c68da7Check integrity

Revisions

2026-10-26 22:40Fact · Supported by the evidence · CurrentRevised the receipt before it
2026-10-15 23:50Fact · Supported by the evidence · CurrentFirst receipt

Claim 2

Class: Fact

Stated directly by a source, with the passage located.

Check: Completed
Assessment: Supported by the evidence
Still current: Current

Independence0 origins confirmed distinct · 1 not shown to be independent · 1 pieces of evidence

Receipt digestsha256:a6de4b91…182adaCheck integrity

Revisions

2026-10-26 22:41Fact · Supported by the evidence · CurrentRevised the receipt before it
2026-10-26 19:22Fact · Supported by the evidence · CurrentFirst receipt

Claim 3

Class: Fact

Stated directly by a source, with the passage located.

Check: Completed
Assessment: Supported by the evidence
Still current: Current

Independence0 origins confirmed distinct · 1 not shown to be independent · 1 pieces of evidence

Receipt digestsha256:cb740c0f…11c62cCheck integrity

10 more claims are in the record.

How it looked when first published is older than the 7 days your plan can replay. See plans

A receipt is not a certificate of truth. It shows how AiSeek assessed a claim and that the record has not been altered since. How receipts work