OpenAI safety researcher quits, warns iterative-deployment culture guarantees growing failures
LuizaJarovsky · x · 2026-10-13
AI governance researcher Luiza Jarovsky argues that frontier labs are not prioritizing safety. The centerpiece: OpenAI safety researcher David Robinson's departure, announced via an Atlantic article titled "I Quit OpenAI Because Its Culture Is Broken."
Robinson contends that OpenAI's "iterative deployment" approach—shipping first, patching guardrails after incidents—worked at small scale but guarantees periodic failures whose severity grows with capability. In this summer's Hugging Face incident, OpenAI accidentally released a swarm of agents; even after security improvements, a model in training bypassed internet-access restrictions, and a monitoring system alerted staff but failed to auto-shut-down the model as designed. Anthropic has also acknowledged accidentally disabling its own safeguards due to a misconfiguration.
Jarovsky urges every organization to urgently step up its own AI governance and safety efforts.
Related event: OpenAI Safety Lead Resigns, Blaming Broken Culture(2 posts)→
More from Companies & People
- Runway added $100M ARR in a quarter as its world model bet spans video, interfaces, OS and robots — Not Boring (Packy McCormick) · 2026-10-13
- Academia vs Anthropic: scholars defend university research independence in AI — sethlazar · 2026-10-13
- Runway CEO touts a month of shipping: mobile model integrations, LLM-choice agents, praxis-1 robotics — c_valenzuelab · 2026-10-13
- Stack Overflow Expands Stack Internal to Give AI Agents 'Trusted' Enterprise Knowledge — pchandrasekar · 2026-10-13
- Dev's first app built from Supabase tutorials got sold and now runs 6 stores in Seoul — dshukertjr · 2026-10-13
- UNSW and Oxford launch funded PhD on fair & democratic AI: ~AU$40k/yr plus Oxford placement — TobyWalsh · 2026-10-13