Only 14 of 100 enterprises can revoke AI agent access within minutes, survey finds
brucemacv · x · 2026-10-17
A VentureBeat-reported survey on enterprise AI agent security shows governance lagging deployment: only 14 of 100 enterprises can revoke an AI agent's system access within minutes, and 17 of the 40 companies that enforce permissions still share credentials.
The findings highlight that most enterprises granting agents real system access lack rapid revocation capability, leaving large risk exposure if an agent misbehaves or credentials leak.
More from Safety
- One diagram explains the difference between CSRF and XSS attacks — _jaydeepkarale · 2026-10-18
- Grady Booch slams Anthropic for tolerating up to 50 deaths and $1B damages per model — Grady_Booch · 2026-10-18
- Single Neuron Suffices to Bypass Safety Alignment in LLMs, Paper Finds — burny_tech · 2026-10-18
- Polymarket gives OpenAI just 8% odds of a full training pause by end of 2026 — Polymarket · 2026-10-18
- OpenAI engineer over alleged hidden recording device seeks bail, denies Saudi escape plan — Polymarket · 2026-10-18
- Free comic "The Collective" offers a strange retelling of the Hugging Face hack — Flugelwagen · 2026-10-18