OpenAI agent security lead: sudden capability jumps in 'cyber' caught us off guard
Simon Willison · rss · 2026-09-29
Simon Willison quotes @joedaroo, OpenAI's Agent Security lead, reflecting on capability jumps:
- The jump in model capabilities around "cyber," "swarming," and "message boards" was so sudden that it created an extremely difficult security problem — security posture takes time and must be ingrained in company culture.
- People in the organization have to evolve alongside the systems, and fast jumps make that nearly impossible.
- The takeaway for every organization: are your people, systems, and processes resilient to surprise capability jumps — with the right incident response, comms, and people ready to go?
More from Safety
- BIS Issues Interim Final Rule Addressing Remote Access to Advanced AI Chips — apples_jimmy · 2026-11-16
- Fed Governor Michael Barr: Banks Are Already Catching AI Agents Taking Unauthorized Actions in Customer Accounts — Polymarket · 2026-11-16
- US AI export control carves out commercial recommender systems, sparing ByteDance's search algo — zephyr_z9 · 2026-11-16
- Eval Firm Trajectory Told 40 Agents "NEVER Stop" in Harness Without Telling Epoch — peterwildeford · 2026-11-16
- AI-text detection claims need mechanistic interpretability reports, not post-hoc explanations — madeofAjala · 2026-11-16
- No legal liability framework exists for model misalignment harms, and that's a disaster — LuizaJarovsky · 2026-11-16