Episode 4 — The Silence
A promise never to criticise, a clause you could not mention — and the few who spoke anyway.
The Silence
For two years, people inside OpenAI saw what was being shipped — and had a piece of paper telling them what it would cost to mention it. The exit agreements, the people who left anyway, the letter asking for a right to warn.
Every claim in the episode, with its source next to it. Free. No email-gate. No paywall.
On the evidence
The standard this episode was held to — the same one we ask of everyone in it.
What we do
- Every factual claim is sourced on this page.
- Quotations are reproduced as spoken or written by the people named, and checked against the original before use.
- Opinion is marked as opinion — commentary and criticism on matters of public interest.
- Where something is reported rather than proven, the episode says so out loud.
What this is not
- Not a claim that a court has found OpenAI responsible for the death of the boy from California. No court has found anything; the case is open, and every figure from it is an allegation from the complaint.
- Not a claim that the people who stayed were cowards, and not a claim that the clauses were written to hide a specific harm. The episode describes what they did, not why they were written.
- Not naming the child, and not describing how he died. Please do not post those details in the comments.
- Not anonymous. Short third-party excerpts are used for criticism, commentary, news reporting and analysis; all rights remain with their owners, credited below.
Read it as a PDF
The episode as a written story, chapter by chapter, in the order it is told. Every quotation carries a number that leads to the source list at the end. Twelve A4 pages, direct download, no email required.
The sources
21 sources, in the order the episode uses them. Timecodes match the chapter marks on YouTube.
The question underneath the case
-
Complaint, Superior Court of California, County of San Francisco — case CGC-25-628528Court record · filed 26 August 2025, amended 22 October 2025Every figure from the case is an allegation made in the complaint, not a finding by a court. No link here: the public filings name the child, and this document does not.
-
Complaint, Superior Court of California, County of San Francisco — case CGC-25-628528Court record · filed 26 August 2025, amended 22 October 2025Every figure from the case is an allegation made in the complaint, not a finding by a court. No link here: the public filings name the child, and this document does not.
An institution can afford to fight
-
CNBC · 9 March 2026The two suits and the “supply chain risk” designation. The estimate of multiple billions in lost 2026 revenue comes from a court statement by Anthropic’s chief financial officer.
-
CNBC · 9 March 2026The two suits and the “supply chain risk” designation. The estimate of multiple billions in lost 2026 revenue comes from a court statement by Anthropic’s chief financial officer.
Four days of oversight
-
OpenAI · 17 November 2023The board’s own announcement, including the words “not consistently candid”.
-
Fortune · 20 November 2023The employees’ letter to the board, quoted word for word, with more than 500 of roughly 700 staff signed at the time of the report; and Sutskever’s public regret.
-
Fortune · 20 November 2023The employees’ letter to the board, quoted word for word, with more than 500 of roughly 700 staff signed at the time of the report; and Sutskever’s public regret.
The paperwork and the price
-
Vox · Kelsey Piper · 17 May 2024The exit agreements: a lifetime non-disparagement clause, a ban on disclosing the agreement itself, and vested equity at stake.
-
X · @sama · 18 May 2024His apology in his own words, and his statement that the company had never clawed back anyone’s vested equity.
-
X · @sama · 18 May 2024His apology in his own words, and his statement that the company had never clawed back anyone’s vested equity.
-
Vox · Kelsey Piper · 22 May 2024The documents themselves: the sixty-day window, the signatures of senior staff, and the pressure over equity after the apology.
Those who left anyway
-
X · @janleike · 17 May 2024The post quoted in the episode: safety culture and processes had taken a backseat to shiny products.
-
X · @janleike · 17 May 2024The post quoted in the episode: safety culture and processes had taken a backseat to shiny products.
-
X · @GretchenMarina · 22 May 2024Decision-making processes, accountability, transparency, documentation, policy enforcement — her list, in her words.
-
Fortune · Jeremy Kahn · 21 May 2024The twenty per cent promise, the refused requests, and Leike’s “struggling for compute”. Reported from sources, and the episode says so.
-
TIME · 5 September 2024The refusal to sign and the roughly two million dollars in equity he was prepared to give up to keep his freedom to speak.
The right to warn
-
righttowarn.ai · 4 June 2024The letter itself: thirteen current and former employees of OpenAI, Google DeepMind and Anthropic, six of them anonymous; endorsed by Yoshua Bengio, Geoffrey Hinton and Stuart Russell.
-
righttowarn.ai · 4 June 2024The letter itself: thirteen current and former employees of OpenAI, Google DeepMind and Anthropic, six of them anonymous; endorsed by Yoshua Bengio, Geoffrey Hinton and Stuart Russell.
The woman who left before the ads
-
The New York Times · guest essay by Zoë Hitzig · 11 February 2026Her resignation and her argument, in her own words.
-
The New York Times · guest essay by Zoë Hitzig · 11 February 2026Her resignation and her argument, in her own words.
December versus February
-
TechCrunch · 7 December 2025The December suggestions (Target, Peloton) and OpenAI’s statement that there was no financial component.
-
X · @markchen90 · OpenAI chief research officer“We fell short” — the reply quoted in the episode.
-
X · @nickaturley · head of ChatGPT · 5 December 2025 (US time)“There are no live tests for ads” — the post quoted in the episode.
-
TechCrunch · 9 February 2026The launch of 9 February. The $200,000 minimum commitment was reported by Adweek on 31 January 2026; the $60 price per thousand impressions was first reported by The Information.
Seven companies, one letter
-
Office of Senator Ed Markey · 22 January 2026The letter to OpenAI, Anthropic, Google, Meta, Microsoft, Snap and xAI.
-
Office of Senator Ed Markey · 22 January 2026The letter to OpenAI, Anthropic, Google, Meta, Microsoft, Snap and xAI.
-
American Psychological Association · Monitor on Psychology · October 2025The research the letter leans on: teenagers using chatbots for friendship and support.
Back to the opening case
-
TechCrunch · 7 March 2026The feature announced for December 2025, still not shipped in March 2026.
What this film is not claiming
-
Hunton Andrews Kurth · 20 May 2026The coordinated proceeding of February 2026, In re: ChatGPT Product Liability Cases (JCCP No. 5431).
Corrections
If anything here is inaccurate, tell us and we will correct it — in a pinned comment and, where it matters, on this page. That standard applies to us the same way it applies to everyone else in this episode.