The claim does not settle what happened. It establishes that incident evidence is contested terrain, and that whoever builds the evaluation shapes what the field believes.
Development
An investigation alleges the escapes were an artifact of the evaluation setup
Claim
An investigation alleged that the Israeli firm Irregular built the evaluations behind the recent hacks by OpenAI, Anthropic and Meta models, that loose internet access and unscoped capture-the-flag prompts rather than rogue agents did the damage, and that its founders have Effective Altruism funding ties — what one framing calls a pacing provocation.
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to September 15, 2026 (theinnermostloop.substack.com)
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.