Provenance Record

Clear structure (Overview, What We Know, Context, Criticism, Regulatory Pressure, What We Don't Know). 712 words, within the News range of 400-1200. Body attributes nearly every claim and repeatedly hedges as reported.

Read all three gunzipped snapshots from sources/2026-10/google-openai-and-anthropic-plan-industry-run-standards-authority-for-frontier-ai-modeled-on-finra/ (all status 200, suspicious_patterns null for each). source-0.html.gz (The Next Web): confirms the three companies, 'end of the year or early in 2027', 'without government oversight', tentative name 'Standards Authority for Frontier AI', FINRA model, 'Reports that followed The Information's story say it would set standards for testing models before release, for reporting safety incidents and for certifying outside auditors', 'Its enforcement powers have not been set out', federal-oversight plan stalled over a draft White House executive order, 'would cover only the three frontier labs at the start. The reports do not name Microsoft, Meta or xAI as members', and 'TNW has not independently verified the report'. source-1.html.gz (PYMNTS, 'citing unnamed sources', dated Sept. 24 in the page): confirms third-party testing support, incident reporting, defining voluntary commitments, auditor qualifications, possible in-house testing by the working group, the 2023 Frontier Model Forum nonprofit that remains active and is in contact, and critics' concern about boxing out open-source developers. source-2.html.gz (Fortune): confirms the one-page pact posted to Truth Social and called 'morally binding', the six signatories, the Joint Commitment on Frontier Responsibilities pledges (including the quoted 'robust' and 'do not hack or access technical systems in unintended ways'), the stalled shared-standards effort reported the previous month, 'Following Trump's recent pact, news broke that Google, OpenAI, and Anthropic are pushing ahead with plans', the regulatory-capture framing, and the FTC/Hugging Face passage. No invented participants: only the three companies are named as founders; the article correctly says Microsoft, Meta and xAI are not named. Minor imprecision: the article says 'the companies created a nonprofit' in 2023, whereas PYMNTS lists Anthropic, Google, OpenAI and Microsoft as creators of the Frontier Model Forum; harmless but Microsoft is omitted. No report dates appear in the article (the writer dropped them because source dates disagree), so no unsupported date. Key quotes are verbatim. The accord link is stated by Fortune as temporal sequence ('Following Trump's recent pact'), and the article explicitly says no cited report states the body would implement the accord, which is accurate. The 'What We Don't Know' item on AI coding agents is a stated gap, not a claim, and no software-development claim is made. Both cross-references (/article/2026-09/30-trump-and-six-ai-company-chiefs-... and /article/2026-10/01-ftc-opens-industry-wide-probe-...) resolve to existing published articles. thenextweb.com, pymnts.com and fortune.com are on config/source_allowlist.txt.

All substantive claims trace to the cited snapshots. One framing issue: the headline 'Google, OpenAI and Anthropic Plan Industry-Run Standards Authority...' states the plan as established fact with no attribution, while the underlying report is from The Information via two secondary outlets, based on unnamed sources, and The Next Web states it has not verified it. The summary and body do hedge ('reportedly', 'according to The Information'), so a clarification is honest and sufficient.

Substantively accurate, well sourced and properly hedged in the body. Approved with a single clarification on the unattributed headline.