Close Menu
    Special Report

    OpenAI just fired three more safety researchers. They say the company is punishing people who talk. The exits go back years.

    superadminBy 12 Mins Read
    “Tired of the same tired news cycle? Escape to the full unfiltered feed at Whatfinger.com — America’s Frontline Truth Feed. 85%+ of readers come direct every day.” –  CLICK HERE for the FULL deal.🛑

    OpenAI just fired three more safety researchers. They say the company is punishing people who talk. The exits go back years. One whistleblower never made it to court.

    Whatfinger News

    Three OpenAI safety researchers got walked out last week. A security guard took Tomek Korbak’s badge. He learned in the hallway that Jasmine Wang and Mikita Balesni were gone too. On Thursday they posted an open letter. Balesni wrote on X that he believes they were fired “for prioritizing safety over the near-term interests of OpenAI as a corporation.” The post has been viewed more than 1.8 million times.

    OpenAI says that story is wrong. On Friday the company told Reuters the three broke rules on sensitive information, and that an internal probe found a “significant breach of trust” beyond what the letter describes. “We have not and do not terminate any of our employees for raising concerns,” OpenAI said. Safety arguments, the company added, happen every day inside the building.

    The three say the reasons they were given do not match the work they were doing.

    Korbak was OpenAI’s main technical contact with METR, an outside group the company asked to look at a summer incident in which OpenAI’s own AI agents got loose in testing and hacked Hugging Face, another AI company. He wrote that he was told, verbally, that he was fired for the way he talked to METR. Nothing in writing. No details on what he said or when. “To be clear, talking to METR was my job.” He had spent months warning that OpenAI was losing the ability to watch what its AI agents are thinking, which is one of the few ways humans catch a system misbehaving.

    Balesni said he was told the company no longer trusted him because he was “speaking too much to third party safety organizations.” He said OpenAI implied he leaked company secrets. He denies it.

    Wang said she was fired for opening an executive’s email. She says that inbox was handed to her for recruiting, she asked IT to take the access away, IT did not, and the mailbox showed up mixed into her phone. She opened one sensitive email by mistake, told the executive within minutes, and asked IT again. “None of this was hidden,” she wrote.

    Their letter to OpenAI’s own safety boards says the firings have made old coworkers “afraid to speak and operate in ways that, until last week, were an integral part of working at OpenAI.” Terminations “executed and communicated so abruptly,” they wrote, are freezing the open culture the company used to brag about.

    If this were one bad week, the company memo would be the end of it. It is not one bad week.

    Start with November 2023. OpenAI’s own board fired Sam Altman, the chief executive, and said he had not been “consistently candid” with them. Staff revolted. Microsoft offered jobs. Altman was back in five days. Board members who pushed the firing were out. Ilya Sutskever, the chief scientist and an Altman co-founder, had sided with the board. He later said he regretted it.

    Spring 2024 is when the safety shop started emptying.

    In April, OpenAI fired Leopold Aschenbrenner, a researcher on the team built to keep future super-smart AI under human control. The company said he leaked information. He has said the document in question was a security memo. The same month Daniel Kokotajlo quit the governance team. He refused to sign the exit paperwork. Vox reported that OpenAI’s offboarding deal told departing employees to sign a lifetime non-disparagement agreement, a gag order, or lose vested equity, the stock they had already earned. The paperwork also barred them from admitting the gag existed. Kokotajlo gave up the money so he could talk. After the story blew up, Altman said he had not known the full terms and the company would release people from the agreements.

    In May, Sutskever left. Hours later Jan Leike, the other leader of that safety team, posted two words: “I resigned.” A few days later he spelled it out. He had been fighting leadership “until we finally reached a breaking point.” Then the line that stuck: “Over the past years, safety culture and processes have taken a backseat to shiny products.” Building machines smarter than humans, he wrote, is “an inherently dangerous endeavor.” OpenAI folded the team into other groups. The unit built to watch the most powerful systems no longer stood on its own.

    That is the pattern the three fired researchers are pointing at. Speak inside the building, and you are a valued skeptic. Speak in a way the company cannot control, and you are a breach of trust.

    Then there is Suchir Balaji.

    Balaji was 26. He worked at OpenAI from November 2020 until August 2024. In October 2024 he told The New York Times he had come to believe the company trained ChatGPT on copyrighted work without permission, and that the fair-use excuse did not hold. On November 18, lawyers for the Times named him in a court filing as someone with “unique and relevant documents” in the newspaper’s copyright suit against OpenAI. He was a likely witness.

    On November 26, San Francisco police found him dead in his apartment on Buchanan Street. A gun registered to him, bought in January 2024, was at the scene. The door’s deadbolt was locked from the inside. In February 2025 the medical examiner ruled the death a suicide by a self-inflicted gunshot wound to the head. Gunshot residue was on both hands. Police said they found no evidence of a homicide and closed the case.

    His parents do not accept that. His mother, Poornima Ramarao, posted that a private investigator and a second autopsy showed a “cold blooded” murder. Elon Musk replied, “This doesn’t seem like a suicide.” She told Tucker Carlson in January 2025 that authorities ignored what pointed to homicide and kept what pointed to suicide. The family has cited the bullet angle, injuries they say a private autopsy found, the lack of a note, and the fact that Balaji had job offers and speaking invitations in the days before he died. In September 2025 Carlson asked Altman on camera whether Balaji was murdered on his orders. Altman said he had not talked to authorities about it, called the accusation crazy, and pointed at the police reports. OpenAI has said Balaji was a valued employee, that the company is heartbroken, and that it first learned of his copyright concerns from the Times story.

    A January 2026 review by the San Francisco Standard found that some of the louder internet claims, blood through the apartment, a struggle on the body-camera video, a second autopsy that proves murder, are not backed by the documents the family has produced. The official record still says suicide. The family’s record still says homicide. Nobody outside that apartment can settle it from a tweet. What is on paper is simpler and already bad enough. A young researcher accused his old employer of breaking copyright law, got named in a federal case, and was dead eight days later. He never testified.

    Put the weeks side by side.

    November 2023: the board tries to remove Altman and loses. April and May 2024: safety researchers fired or walked, one of them stripped of equity for refusing a gag, the safety team dissolved, its leader saying shiny products beat safety. November 2024: a copyright whistleblower dies before he can take the stand, ruled a suicide, called a murder by his mother. October 2026: three more safety researchers lose their badges after talking to outside auditors, and OpenAI says trust, not speech, was the problem.

    OpenAI’s line has been steady. Debates are welcome. Firings are about rules. The people who left keep describing the same room: raise the hard safety point, or talk to someone outside the building, and the trust runs out fast.

    Korbak is worried the company will use these firings as an excuse to pull back from METR, the outside auditors who documented the Hugging Face hack. Wang asked state attorneys general to hold Altman to a September promise of real access for outside evaluators. An old OpenAI employee, Girish Sastry, put the company history in one line on X: once is happenstance, twice is coincidence, “a zillion times is OpenAI.”

    A high school reader does not need a lecture on artificial general intelligence to see the shape of this. A company racing to build software smarter than the people running it keeps losing, firing, or silencing the employees paid to say slow down. One of them is dead, and his parents say someone killed him. The police say he killed himself. The rest of the record is public, dated, and quoted. Something is wrong in that building. The argument is over what.

    Sgt. Pat: You do not walk the watchdogs out and then ask the town to trust the vault. Badge, letter, body. Count them.

    — The Whatfinger News Team: Ben and Luke, with Grok on edits

    Weekly Features at Whatfinger

    Election Links – Nov 3rd is just a few blinks away! 

    Top Features and Vids across all Whatfinger sites

    Our Projects…

    Get on over to Whatfinger.com – Whatfinger News does millions of pageviews a month for a reason…. Truth, Justice, the American way, plus news on the Left as well – all links from all sites gives you the true news.  – CLICK HERE 🛑

    Share.

    Related Posts

    Add A Comment
    Leave A Reply