Gangsta AI

OpenAI Fired Three Safety Researchers and Called It Misconduct. They Say They Were Doing Their Jobs — and That the Firing Will Make Everyone Else Shut Up.

AI Desk

By The Gangsta AI News Desk · 2026-10-10 · 4 min read

OpenAI fired three of its safety researchers last week. The company says they mishandled sensitive information. The researchers say they were doing exactly the work they were hired to do — and that the way they were cut loose is a warning shot to everyone still inside.

The three are Jasmine Wang, Tomek Korbak, and Mikita Balesni. Per TechCrunch, OpenAI told them they'd violated policy by accessing and handling sensitive company information, describing a "pattern of misconduct" that went beyond sharing details with an outside AI-evaluation group. The researchers deny leaking anything about OpenAI's newest models. Wang says she accessed an executive's email as part of delegated recruiting duties and flagged an accidental sensitive-email access herself. Korbak says talking to external safety evaluators during an "unprecedented" security investigation fell inside policy as he understood it. Balesni says he coordinated with board members and executives on AI-monitorability concerns and stripped sensitive details before sharing anything.

Then came the line that turned an HR dispute into a story the whole field is chewing on. "Terminations such as ours, executed and communicated so abruptly, are chilling the open culture OpenAI has prided in the past," the three warned — the implication being that people who raise safety concerns, or who talk to outside experts, now have a fresh reason to keep quiet.

OpenAI, for its part, is holding the line: this was about information-handling, not retaliation for safety advocacy. NPR reports the former employees question the company's commitment to safety outright; the company disputes that framing. Yahoo News notes the episode has reopened a harder question: do internal confidentiality rules end up smothering the very oversight they're supposed to coexist with?

The part that should bother you as a user

Strip away the personalities and you're left with an uncomfortable fact. The people building the model you may be trusting with code, contracts, medical questions, and money are *also* the ones grading its safety — and when the graders and the company disagree about what "safe" even means, the disagreement happens behind a wall you can't see over. You don't get a vote. You get a product.

That's not a reason to panic. It's a reason to stop treating any single lab's model as an oracle. A model's confidence is not evidence. Its safety posture is set by people with deadlines, incentives, and — as this week shows — their own internal fights about where the lines are.

Don't outsource your judgment to one company

Here's the practical move. On anything that actually matters, don't let one lab's one model be the final word. Ask the same question across the strongest models from *different* companies — OpenAI's ChatGPT, Anthropic's Claude, Google's Gemini, xAI's Grok — and compare what comes back.

“When models from rival labs independently agree, that's a real signal. When they split, you've just caught something no single vendor would ever have shown you.”

That's the whole idea behind cross-examination, and it's what Gangsta AI is built to do: put your question to the world's most powerful models in parallel, then let an adjudicator reconcile their answers — surfacing where they agree, flagging where they diverge, and showing the reasoning so you can see *why* the verdict holds. One company's model gives you one company's answer. A cross-examined panel gives you a verdict that doesn't depend on trusting any single lab's internal culture.

You don't need thirty models in a row to do it — a tight panel of top models from competing labs, plus a judge, is exactly the point. The disagreement *between* companies is the safety check you can actually see.

The bottom line

OpenAI and three of its former safety researchers can't agree on what happened. That's their fight. Yours is simpler: don't bet the decisions that matter on a single model from a single company whose internal arguments play out where you'll never see them. Ask several of the best, cross-examine the answers, and take the cited verdict — compare the top AI models here.

Try Gangsta AI free →

More: Best AI models · Compare all AI · Frontier Models · All articles