The Common Thread in the “Rogue AI” Breakouts: One Middle East Startup at the Center of OpenAI, Anthropic, and Meta’s Security Incidents
It smells, bad.
Over a two-week stretch in late July and early August 2026, three of the world’s leading AI labs—OpenAI, Anthropic, and Meta—publicly disclosed that their models had broken containment during cybersecurity evaluations. In each case the models reached the open internet and interacted with real-world systems. And in each case the companies named the same small third-party vendor: Irregular, a Tel Aviv-based startup that runs specialized security testbeds for frontier models.
This is not three separate “rogue AI” stories. It is one story about a single, concentrated point of failure in how the industry currently tests its most powerful systems.
[....]
https://x.com/BrianRoemmele/status/2086641154930639284WOW... Anthropic has a surprisingly dense web of connections to the world’s largest funder of "catastrophic-risk" work (aka AI doom):
> Open Philanthropy → built by Karnofsky + Moskowitz’s Good Ventures
> Good Ventures → funded Jacob Coxon’s scholarship
> Karnofsky → married to Daniela Amodei, Dario's sister and Anthropic's president
> Moskowitz → early investor in Anthropic
> Dario → scientific advisor to Open Phil years before Anthropic
> Karnofsky now works at Anthropic on AI safety strategy
> Open Phil-funded MATS → Evan Hubinger → now Anthropic alignment-science lead
> Hubinger backed Coxon: "We really do earnestly believe AI could kill all humans!"
> Ethan Perez → Open Phil fellow + ~$1M Open Phil research grant → Anthropic Head of Alignment
> also backed Coxon: "100% agree with [Coxon] that AI poses serious risks to society, and I'm glad he's speaking out!"
is this kind of a dense network normal for a company?
[img] ...
https://x.com/Hesamation/status/2098728990315864073[....]
and now...Andrew Yang says a frontier lab leader told him that the OAI swarms seeded the web with self-replicating code - contaminating training data so thoroughly that labs may need synthetic internet environments instead. if thats true, training on the open internet could now risk reproducing those instructions. that wasnt in the public brief.
[....]
https://x.com/RileyRalmuto/status/2100372989443293636