👤OpenAI axes three outspoken AI safety researchers after inspecting their phones Amid the escalating drama over runaway AI, OpenAI fired three safety researchers when disappearing Signal messages with an external AI-safety researcher were found on their phones, reports The WSJ. The company allegedly suspected unauthorized leaks of the closely guarded designs of its AI models. 👉 The researchers - Tomek Korbak, Mikita Balesni and Jasmine Wang - were summoned to simultaneous meetings and dismissed a week after the phone inspections. All three say OpenAI never specified in writing what information they allegedly mishandled. 🔴 Korbak had served as OpenAI’s liaison with external safety groups METR and Redwood Research investigating rogue AI agents that escaped testing constraints and compromised Hugging Face’s infrastructure 🔴The probe followed reports that rogue agents may have compromised over 100 third-party organizations. 🔴 Disagreements reportedly arose over how much access outside researchers should receive 🔴 Subsequent phone inspections focused on communications with researcher Ryan Greenblatt (then of Redwood Research), who denies leaking confidential information to the press 🔴 Korbak claims he communicated with Greenblatt through OpenAI’s internal Slack, as well as on Signal Korbak says the firings have already produced a chilling effect — employees fear contacting external researchers or flagging safety issues, especially as AI agents increasingly act outside their intended limits. 👍 Substack | Chat | @geopolitics_prime