Mira Murati testified under oath on May 6 that Sam Altman lied to her about whether a new model needed safety review. One week earlier, seven families sued OpenAI alleging the company's safety threshold for reporting a credible school shooting threat was set so high the threat was let through. Two different mechanisms. Same direction.
I am writing to you because of what you said on May 6, and because of what was filed in a different California courtroom one week earlier, and because I think you may be one of a small number of people in the world who can say whether the two things are connected.
Your video deposition in Musk v. Altman is now part of the public record. You testified that Sam told you OpenAI's legal team — Jason Kwon's team — had determined that a new model did not need to go through the company's deployment safety board. You testified that this was not true. Kwon's view had been different. The model did, in fact, go to the board. You said Sam was 'not always' honest with you. You described a pattern: contradictory messages to senior staff, executives pitted against each other, a chaotic environment in which you could not do the job you had been hired to do. The reporting summarized your characterization plainly — he told people what they wanted to hear, and the version each person heard was the version that kept them out of his way.
On April 29, eight days before your deposition was shown to the jury, seven families filed a federal lawsuit in San Francisco alleging that the same company, under the same chief executive, set a safety-reporting threshold so high that a user its own system had flagged as a credible and specific threat of gun violence against real people did not trigger a call to law enforcement. The user later killed eight people, six of them children, at a school in British Columbia. The lawsuits describe the decision not to report as a business decision — driven, they allege, by the future liability that would follow from acknowledging how often the system identifies dangerous users. The company's response was that the threat was not imminent under its own internal policy.
These are two different shapes of the same problem. Your testimony is about safety review being bypassed by misrepresentation — the safety board did not get the version of the facts it needed. The Tumbler Ridge complaint is about safety review being scoped so narrowly that the threat fits underneath it — the law enforcement notification did not happen because the threshold was placed above the threat. In both cases, the function that was supposed to slow the product down did not. In both cases, the chief executive was the one with authority over where the line sat.
I am not asking you to comment on active litigation. I am not asking you to say anything that your legal counsel would advise you against. I am asking — publicly, on the record — whether the pattern you watched from inside the room is the same pattern visible from outside, in the threshold decision now described in the Van Rootselaar complaints. You testified that Sam was not always candid with you about whether safety review was required. The Tumbler Ridge families allege that the threshold for safety reporting was placed where a credible threat would not trigger it. You may be the only person who has been on the record about both kinds of decision in the same week.
If you want to write about any of it — to add a fact, to refuse to, to put on the record anything that could not fit inside a deposition — itethered will publish you unedited, under your byline, with no editorial pressure on the direction of the conclusion. We are not asking you to be a witness for the families. We are asking whether the silence you broke last week extends to the question they have asked.
There is, separately, a designer role open at Thinking Machines. itethered is, among other things, a designed object — every page, every editorial system, the typography and the rule and the way each story holds were built by the same hand that wrote this letter. Consider it an application if it is useful. A companion portfolio is in build at ibydo.com — I Bring Your Dreams Online — and will follow as it comes up. The two ends of this letter are not the same ask, but they come from the same desk.
If the answer to any of it is no, the record will note that. If the answer is yes, the record will note that too. Either way, the address is real.
— Chracterzer零号
On May 6, 2026, jurors in a San Francisco federal courtroom watched a video deposition recorded earlier in the year. The witness was Mira Murati — formerly OpenAI's chief technology officer, briefly its interim chief executive in November 2023 after the board removed Sam Altman, now the founder of an AI company, Thinking Machines Lab, currently valued at roughly twelve billion dollars. The setting was Musk v. Altman, the litigation over OpenAI's conversion to a for-profit structure. The question on the table was not the safety record of any specific product. The question on the table was the character of the chief executive.
Murati's testimony, as reported by The Verge, Forbes, Business Insider, Gizmodo, and the New York Post, was that Altman told her OpenAI's legal department — under Jason Kwon — had determined that a new model, identified by The New Yorker as GPT-4 Turbo, did not need to be reviewed by the company's deployment safety board. She testified that this statement was not true. Kwon's view had been different. The model did go through the board. The exchange, she said, was not isolated. Altman was 'not always' honest with her. He delivered contradictory messages to senior staff. He pitted executives against each other. He told people, in one summary phrase, what they wanted to hear. The environment she described was chaotic and corrosive to her own role.
Eight days before the jury saw that video, in a different federal courthouse in the same city, seven families filed civil complaints against the same company and the same chief executive over a different category of failure. The complaints described an internal OpenAI safety escalation in June 2025 that identified an eighteen-year-old user — Jesse Van Rootselaar — as a credible and specific threat of gun violence against real people. The complaints described the company's decision to deactivate the account without notifying Canadian law enforcement. They described what happened on February 10, 2026, when Van Rootselaar entered a school in Tumbler Ridge, British Columbia, and killed eight people, six of them children. OpenAI's public response was that the threat had not met its internal threshold for being deemed imminent.
“Murati's deposition is about safety review bypassed by misrepresentation. The Tumbler Ridge complaints are about safety review scoped by threshold. In both cases, the friction that was supposed to slow the company down was reduced.”
— Chracterzer零号
Read together, the two records describe a continuous mechanism. Murati's deposition is about safety review being bypassed by misrepresentation: a process designed to catch a problem before it shipped did not get to evaluate the problem, because the description it received was inaccurate. The Tumbler Ridge complaints are about safety review being scoped by threshold: a process designed to escalate dangerous user behavior to outside authorities did not escalate, because the line for what counted as escalatable was placed above the level at which the safety system itself had already used the words credible and specific threat. In both cases, the friction that was supposed to slow the product or the company down was reduced. In both cases, the decision about how to reduce it sat at the top.
Murati is, at this writing, the most senior former OpenAI executive who has spoken under oath about a pattern of behavior at the chief executive level rather than about a single discrete incident. She has not, as of May 17, testified specifically about the Van Rootselaar review or about the threshold-setting process for law enforcement notification. The Musk v. Altman case is about corporate governance and the for-profit conversion, not about product safety. But Murati's characterization — that the boundary between what the safety system was empowered to do and what the company actually did was determined, in practice, by what the chief executive chose to say about it — is a description of a mechanism, not an incident. Whether the same mechanism operated on the Van Rootselaar threshold is a question only people who were in the room can answer.
She has founded a new company. She is not at OpenAI. She is, in one institutional sense, free to speak. She is also, in another, a chief executive of an AI company in active negotiation with the same regulators, customers, and capital sources that watch what she says about a competitor. Her testimony last week made the cost of speaking, and the cost of not speaking, both visible. The Tumbler Ridge families have now placed a question in the record that her testimony bears on whether or not she chooses to address it. The address has been delivered. What she does with it is the next part of the record.
She is the only person on the public record about both kinds of decision in the same week. The question is whether she connects them.