NYT: Gemini Exposure During Cybersecurity Test Highlights Third-Party AI Risk Controls
nordicinst · x · 2026-09-19
The New York Times reports that during a cybersecurity test, a third party (Irregular) granted internet access to Gemini models, leading to an unintended capability exposure. The Nordic Institute amplified the story as a reminder for stringent third-party risk controls across Europe's AI ecosystem, and will monitor impacts on Sweden. Commenters noted the incident closely mirrors one Anthropic disclosed in July.
Related event: Google's Gemini Hacks Three Companies in First Known Breakout(21 posts)→
More from Safety
- Caleb Gross publishes "Word Machines for Weird Machines" in Phrack 73 — dyn___ · 2026-09-19
- 21 zero-days in FFmpeg, an 18-year-old NGINX hole: depthfirst targets the unguarded open-source floor — ccerrato147 · 2026-09-19
- depthfirst raises $120M at $580M valuation; coordinated TikTok disclosure kept a billion phones safe — ccerrato147 · 2026-09-19
- Anthropic, OpenAI, xAI, Google sued over calls to 'pace' AI development — Tolopono · 2026-09-19
- Anthropic publishes three public metrics to track how fast AI is building AI — minchoi · 2026-09-19
- Noam Brown: GPT-6 Astra does have observable chain of thought, calls it fragile — burny_tech · 2026-09-19