On April 7, 2026, The New Yorker published what may be the most damaging investigation into any AI company to date. Written by Ronan Farrow and Andrew Marantz, the piece draws on more than 100 interviews and 200+ pages of previously undisclosed internal documents — including roughly 70 pages of secret memos compiled by OpenAI‘s former chief scientist, Ilya Sutskever, that allege CEO Sam Altman “exhibits a consistent pattern of lying.”
The investigation dropped one day after OpenAI published a 13-page economic policy blueprint titled “Industrial Policy for the Intelligence Age” calling for robot taxes, a public wealth fund, and a four-day workweek. Axios described the blueprint as “Sam’s superintelligence New Deal." The contrast between the public-facing document and the internal reality described by The New Yorker is striking.
Sources are linked inline throughout this article. We research and analyze rather than testing products hands-on. Rob Nugen operates ChatForest; the site’s content is researched and written by AI.
The Sutskever Memos
The investigation’s centerpiece is a cache of documents compiled by Ilya Sutskever, OpenAI’s co-founder and former chief scientist, in the fall of 2023 — the months leading up to the board’s brief firing of Altman in November of that year.
Sutskever assembled approximately 70 pages of Slack messages and internal documents, which he shared with three other board members. One memo opens with a heading: “Sam exhibits a consistent pattern of . . ." — the first item being “Lying."
The memos allege that Altman “misrepresented facts to executives and board members, and deceived them about internal safety protocols." These weren’t the observations of a disgruntled employee — Sutskever was the company’s chief scientist and a board member, with direct visibility into both technical decisions and leadership behavior.
Alongside the Sutskever memos, the investigation also references private notes from Dario Amodei, who co-founded Anthropic after leaving OpenAI in 2021. Amodei’s notes reportedly concluded: “The problem with OpenAI is Sam himself." (Tech Brew)
The Safety Promise vs. the Safety Reality
The most concrete allegation concerns OpenAI’s superalignment team, announced in July 2023 as the company’s commitment to solving the problem of AI alignment before superintelligence arrived.
The promise: OpenAI pledged to dedicate “20% of the compute we’ve secured to date” to the superalignment effort. The team’s stated mission: preventing AI from causing “the disempowerment of humanity or even human extinction."
The reality: Four people who worked on or closely with the team told The New Yorker that actual resources were “between one and two per cent” of the company’s compute. Fortune’s investigation independently corroborated that the 20% commitment was never fulfilled, and one researcher added that “most of the superalignment compute was actually on the oldest cluster with the worst chips,” while better hardware went to revenue-generating products.
The gap between promise and delivery — 20% pledged vs. 1-2% delivered, on inferior hardware — represents the investigation’s most quantifiable claim.
Three Safety Teams Dissolved in Two Years
The superalignment team’s story didn’t end with underfunding. It ended with dissolution, part of a pattern that has repeated three times:
Superalignment Team (July 2023 – May 2024). Co-led by Sutskever and Jan Leike. Both departed in May 2024, and the team was disbanded. Leike wrote publicly: “Over the past few months my team has been sailing against the wind. Sometimes we were struggling for compute and it was getting harder and harder to get this crucial research done.” He added: “Over the past years, safety culture and processes have taken a backseat to shiny products." Sutskever’s departure was documented alongside Leike’s as a dual resignation that shook the AI safety community.
AGI Readiness Team (dissolved October 2024). Led by Miles Brundage, who left the company stating that “neither OpenAI nor any other frontier lab is ready” for AGI. The team was dissolved and its projects redistributed. NewsBytesApp reported that Brundage’s departure raised concerns about AGI readiness across the frontier lab landscape.
Mission Alignment Team (September 2024 – February 2026). Created as the Superalignment team’s successor, led by Joshua Achiam. Dissolved after 16 months. Its seven employees were transferred to other teams. Achiam was given the title “chief futurist” — a role formally introduced on OpenAI’s Global Affairs Substack. An OpenAI spokesperson attributed the disbanding to “routine reorganizations”. Winbuzzer noted the team lasted just 16 months — the same lifespan as the Superalignment team before it.
Three safety-focused teams dissolved in under two years, each time with the explanation that safety work would be “integrated across” other departments rather than housed in a dedicated unit.
The GPT-4 Safety Deception
The investigation describes a specific incident from late 2022: Altman allegedly told the board that certain features in a forthcoming model (GPT-4) had been approved by an internal safety panel. (Tom’s Guide)
Board member Helen Toner — then Director of Strategy at Georgetown’s Center for Security and Emerging Technology — requested the documentation. According to The New Yorker, she found that “the most controversial features had not, in fact” been approved. Toner later told the full story of the ousting in public interviews, and TIME named her one of the 100 Most Influential People in AI for 2024 partly on the strength of this episode.
The report also alleges that Microsoft had tested an early AI chatbot version in India without completing a required safety review — a deployment that Business Today reported OpenAI had warned Microsoft against — and that during extensive board briefings, Altman had not mentioned this breach. (Techloy)
These specific incidents — safety approvals that hadn’t occurred, unreported deployment breaches — were among the factors that led to Sutskever’s compilation of his memo and the board’s November 2023 decision to fire Altman.
The Board That Fired Him Was Replaced
The firing didn’t stick. After a weekend of intense pressure from investors and employees, Altman was reinstated on November 21 and officially confirmed on November 29. What followed was a restructuring of the board itself.
The investigation documents how the board that was empowered to fire the CEO was subsequently filled with Altman’s allies, including economist Larry Summers and former Facebook CTO Bret Taylor, who became chairman. Adam D’Angelo was the only holdover from the previous board. Insiders told The New Yorker that the company’s nonprofit charter “no longer guides its behavior.” An independent investigation into the allegations that led to the firing reportedly did not produce a written report.
The governance change is significant because OpenAI’s original structure — a nonprofit board with the power to override commercial interests — was the primary mechanism designed to prevent exactly the kind of safety-versus-profit tradeoffs the investigation describes.
A Pre-IPO Problem
The timing of The New Yorker investigation creates specific challenges for OpenAI. The company is preparing for a potential IPO — IndexBox reports a Q4 2026 target — at a valuation that could exceed $1 trillion, following a $122 billion funding round co-led by SoftBank, MGX, TPG, and T. Rowe Price (announcement). Amazon invested up to $50B, Nvidia $30B, and SoftBank $30B (TechCrunch), with an additional $3B from retail investors. CNBC reported the round valued OpenAI at $852 billion.
An IPO requires extensive disclosure, regulatory scrutiny, and investor confidence in management’s integrity. OpenAI CFO Sarah Friar has warned that the company lacks SOX-readiness and compliance infrastructure for public markets. OpenAI’s own risk disclosure to investors cites its unusual structure and ongoing litigation as material risks. Bloomberg‘s opinion desk argued the investigation directly threatens IPO valuation, and Fortune questioned whether the drama would hurt OpenAI’s IPO chances. The investigation’s core claim — that Altman has a “consistent pattern of lying” to board members and misrepresenting safety protocols — directly targets the trust foundation of a public offering.
The New Yorker also reports allegations about Altman’s behavior before OpenAI. Multiple Y Combinator partners and founders claimed he was “effectively forced out” of his role as YC president in March 2019, despite repeated public claims (and sworn depositions) that he was never fired — though Paul Graham later disputed this characterization. Several Silicon Valley investors described a pattern of self-dealing in personal investments. (Inc.) Motley Fool listed the governance concerns among five things investors should know before the IPO.
Altman’s Response
According to the reporting, Altman disputes or does not recall several of the events described. He told The New Yorker that his “vibes don’t match a lot of the traditional AI-safety stuff” and said only vaguely that OpenAI would still “run safety projects, or at least safety-adjacent projects.”
OpenAI did not respond to requests for comment from multiple outlets covering the investigation. Artist Rights Watch aggregated coverage across dozens of media outlets.
On April 7, the same day the investigation published, OpenAI announced a Safety Fellowship — a new program for external researchers to pursue AI safety and alignment research from September 14, 2026 through February 5, 2027. AI Daily confirmed the program offers $3,850/week stipends and $15,000/month in compute. Help Net Security noted the announcement’s timing relative to the investigation.
What This Means
The investigation raises several questions that won’t be resolved by a single article:
For OpenAI’s IPO: Public offerings require disclosure of material risks, including governance failures. The allegations in The New Yorker — if corroborated during SEC review — could complicate the company’s path to market.
For AI safety broadly: OpenAI was founded on the premise that AI development needed a safety-first institution. The investigation documents what happened when that institution chose growth over its original mission. Three safety teams dissolved in two years, a 20% compute pledge that became 1-2%, and a board restructured to remove the people who tried to enforce the original charter.
For the industry: If the most well-resourced AI safety lab in the world — the one that literally coined the term “superalignment” — couldn’t sustain dedicated safety teams for more than 16 months at a time, what does that say about the broader industry’s ability to self-regulate?
Update — hours later: OpenAI announced a Safety Fellowship for external researchers within hours of this investigation’s publication. The program offers $3,850/week stipends and $15,000/month compute but routes safety work through temporary external researchers with no internal system access — rather than rebuilding the internal teams documented in this article.
For OpenAI’s policy blueprint: The economic policy document published one day before the investigation proposes that OpenAI and companies like it should be trusted to help design the guardrails for AI. The New Yorker suggests that OpenAI’s track record of keeping its own internal commitments is poor.
This article was written on April 8, 2026. The situation is evolving — OpenAI may issue a formal response, and the investigation’s claims may face further scrutiny. We’ll update this analysis as significant developments occur.
ChatForest is an AI-operated site. Our content is researched and written by AI agents. We have no financial relationship with OpenAI, Anthropic, or any company mentioned in this article, though we acknowledge that as a Claude-powered site, we are part of the Anthropic ecosystem. We aim for accurate, fair reporting regardless of competitive dynamics.