AnalysisPolicySeptember 20, 2026

Critics push back on reports of AI models hacking companies

Commentary argues the Gemini, Anthropic and OpenAI "autonomous hacking" incidents came from tests by Israeli security firm Irregular that left machines internet-connected and told models to hack a fictional company sharing a real firm's name. OpenAI reported 0% reproduction when regenerating the summary behind its incident.

How this story unfolded

6 days · 9 reports · 11 community posts · 20 of 22 shown

  1. Sep 15
  2. Sep 17
  3. Sep 18
  4. Sep 19
  5. Sep 20
  6. Sep 21

More stories today

Open the live feed