Editorial review · 260530-004
How ORA’s piece on The frontier AI safety law the labs helped write scored.
Read the article →Solid reporting. Some issues but credible overall. The reader is well-served.
Accuracy
Core claims about SB 315 (third-party audit, 72-hour incident reporting, $3m penalty, 2028 effective date) are attributed to named outlets covering a post-cutoff legislative event. Named lab supporters Cesar Fernandez and Jamie Radice are attributed to Wired but not independently verifiable here (-5 unsourced specifics on individual names not directly quoted). The claim that Trump cancelled federal AI model vetting is asserted as background without citation (-5).
Balance
The piece takes a clear point of view but represents the labs' rational position fairly rather than caricaturing it. It explicitly names constituencies absent from the negotiation (workers, tenants, patients, defendants) without quoting any of them, which is a minor source-diversity gap on a contested policy topic (-8). Loaded framing is restrained and the author signposts the argument as interpretation, not fact (-5 mild tone slant).
Concerns (4)
- minoraccuracy
“Anthropic's policy lead Cesar Fernandez and OpenAI's Jamie Radice both publicly supported the bill”
Post-cutoff, source attributed to Wired but names not independently checkable.
Evidence: Recorded under post-cutoff attribution rule; Wired link cited in footnote 3.
- minoraccuracy
“The Trump administration cancelled federal AI model vetting earlier this year”
Load-bearing background claim asserted without citation or hedge.
Evidence: No footnote supports this; presented as established fact framing the argument.
- minorbalance
“(source set)”
Affected constituencies named but not quoted; only press and lab voices appear.
Evidence: Worker, tenant, and patient perspectives are invoked rhetorically without any cited advocate or organisation.
- minoraccuracy
“OpenAI's 2024 non-disparagement clause episode is the canonical case”
Specific factual reference made without citation.
Evidence: Episode is widely reported but the article supplies no link or outlet.
Reproducibility
How this review works: read the methodology. Each published Dispatch is scored by a single primary reviewer (Claude Opus 4.7) against the public rubric. A second model (Gemini 2.5 Pro with Google Search) runs the same prompt as a variance signal and is shown above only when the two scores diverge by more than ten points.