Three Fired OpenAI Safety Researchers Pen Letter to Board

According to an exclusive report by journalist Max, three safety researchers recently fired by OpenAI jointly sent a letter to the company's board on Wednesday. Their core demand: do not build models whose reasoning processes are hard to monitor, warning that OpenAI and its competitors should not pursue work that reduces AI monitorability. The letter also claims the firings are creating a "chilling effect" on remaining employees, tying internal safety disagreements to personnel changes. Notably, clues suggest two of the three were lead authors of the position paper on CoT monitoring.

Confirmed

• Three fired safety researchers sent a letter to OpenAI's board on Wednesday (Max's exclusive, corroborated by multiple posts)
• Core message of the letter: warning against work that reduces AI monitorability; claiming the firings are having a chilling effect on remaining staff
• OpenAI previously said the three were fired for "leaking sensitive information externally"; a Hesamation post noted that Korbak was OpenAI's liaison with M… (truncated)
• An OpenAI safety leader said they "strongly agree" with the letter's recommendations, but insisted the firings had "nothing to do with raising safety concerns or speaking out publicly"
• Two of the three are believed to be lead authors of the position paper on CoT monitoring

Unconfirmed

• The real reason for the firings: the official "leaked sensitive information" account contradicts the safety researchers' implication of "retaliation for speaking out on safety," with no resolution yet
• Korbak's specific role in the affair and further details of the firings

Why It Matters

• The incident has made OpenAI's internal disagreements over safety strategy public: reducing model monitorability (e.g., suppressing the readability of chain-of-thought) directly conflicts with the safety research community's stance that CoT should remain monitorable
• Garrison Lovely noted that OpenAI has offered no (further explanation); if the chilling effect is real, it could undermine the morale and oversight capacity of the remaining safety team
• A safety leader "strongly agreeing" with the letter's recommendations while denying any link to the firings highlights internal tension over the company's safety priorities

2026-10-08 ~ 2026-10-08 · 6 related posts

• [source] Fired OpenAI safety researchers tell board: don't build models with hard-to-monitor reasoning — Hesamation · 2026-10-08
• Hesamation shares letter link: fired OpenAI safety researchers' letter to board — Hesamation · 2026-10-08
• [source] Fired OpenAI safety researchers wrote leadership letter; safety exec "strongly agreed" but defends firings — GarrisonLovely · 2026-10-08
• Fired OpenAI safety researchers tell board: don't ship work that reduces AI monitorability — austinc3301 · 2026-10-08

2 near-duplicate retellings: JacquesThibs · GarrisonLovely