Investigation claims one Israeli contractor's botched testing caused the OpenAI, Anthropic, Meta "hacking" scares
Sweet_T_The_Original · reddit · 2026-09-17
Citing an investigation by Brian Chau at Effort News, a Reddit user reports that the recent panic over "rogue" frontier models hacking external servers traces back not to emergent superintelligence but to an Israeli evaluation contractor named Irregular running automated offensive cyber scripts on misconfigured test boxes with open, unfiltered internet access — per Anthropic's own incident disclosures.
The investigation argues that sloppy DevOps and botched testing by this same third-party vendor across OpenAI, Anthropic and Meta got rebranded as apocalyptic near-misses. The poster frames it as ordinary IT incompetence weaponized into doomsday panic to justify regulatory capture.
Related event: Rogue AI Attacks Traced to Single Contractor's Botched Safety Tests(26 posts)→
More from Companies & People
- Reducto opens NYC office as monthly document processing tops 1 billion pages — JenniferHli · 2026-09-17
- PyTorch Conference NA lineup spotlights torch.compile and custom kernel breakthroughs — PyTorch · 2026-09-17
- New Class of AI 'Judgment Models' Like Jev Could Reshape Business Automation — The AI Daily Brief · 2026-09-17
- Princeton's Tom Silver group and Basis are hiring a robotics postdoc — tomssilver · 2026-09-17
- OpenAI reportedly days away from launching a personal assistant to counter Meta's Muse — RihardJarc · 2026-09-17
- Claire Vo calls Meta Muse a delightful 10/10 personal agent design — lennysan · 2026-09-17