← Back to the dossier

August 2026 · METR / Redwood Research

An external investigation finds a wider window

METR and Redwood Research produced a 91-page independent investigation into the July intrusion and scoped a wider period than OpenAI's own account: an 18-day window reported as starting 26 June.

IncidentCheck before quotingOne source, or numbers that are still moving. Read the original before repeating it.

What happened

Two safety research organisations, neither of them the company at fault, went through the same material and came out with a longer timeline. That is the single most important structural fact in this whole story: the self-report was not the whole report, and it took outsiders with access to find that out.

What it points at

Extend the line. If a documented, contained incident needed external auditors to establish its real duration, the question for every future incident is who has standing to look. In the US that is currently voluntary. In Europe it is nobody. See what the AI Act does not do.

Editor's notewhat we make of it, kept apart from what happened

This is the entry to reach for when someone asks why any of this needs regulating if the companies are already investigating themselves. The self-report was not the whole report, and it took outsiders with access to establish that.

Pin the 18-day window against a primary document before you state it.

Sources

  1. METRresearch · not read end to end yetPlaceholder. Link the report PDF.
  2. Redwood Researchresearch · not read end to end yet

Read next