OpenAI Insiders Say Sam Altman Is the Problem, Trust Erodes Within Company

OpenAI’s internal turmoil surfaced in a parallel set of disclosures released on April 6, 2026. The company’s own policy brief, published the same day it announced a slate of “superintelligence‑safety” recommendations, emphasized a “people‑first” approach, pledging transparency around risks such as AI systems evading human control or being weaponized by authoritarian regimes (Ars Technica). Yet, a concurrent investigation by The New Yorker, which interviewed more than 100 current and former OpenAI staff and examined internal memos, painted a starkly different picture of the firm’s leadership dynamics. The magazine’s reporters found that senior engineers—including former chief scientist Ilya Sutskever and ex‑research head Dario Amodei—had documented a pattern of “alleged deceptions and manipulations” by CEO Sam Altman, culminating in a consensus that the CEO was “not fostering a safe environment for advanced AI” (The New Yorker).

The New Yorker’s account highlights two contradictory traits that board members attribute to Altman: a compulsive need to be liked and a “sociopathic lack of concern for the consequences” of his actions. One board member is quoted as saying Altman possesses “two traits that are almost never seen in the same person” – a desire to please coupled with an apparent willingness to deceive when it serves his personal agenda (The New Yorker). This characterization is reinforced by internal communications cited in the investigation, where Amodei wrote plainly, “The problem with OpenAI is Sam himself.” The memo chain, according to the magazine, shows a cumulative erosion of trust that is not tied to any single incident but to a series of strategic decisions that staff perceived as prioritizing Altman’s power over safety protocols.

Altman’s public response to the New Yorker story has been limited to vague denials and claims of “forgetting” certain events, a stance that the investigation describes as “conflict‑avoidant” (The New Yorker). He has also framed the shifting narrative as a reaction to the “changing landscape of AI,” suggesting that policy pivots are driven by external pressures rather than internal mismanagement. Nonetheless, the timing of the policy brief— which includes proposals such as experimenting with shorter workweeks for AI researchers and establishing a public oversight board—appears at odds with the internal sentiment that leadership is “shifting away from positioning OpenAI as a savior” and toward “ebullient optimism” (Ars Technica). Critics argue that this tonal shift may be an attempt to deflect scrutiny rather than address the substantive safety concerns raised by staff.

The broader context amplifies the stakes. OpenAI’s models now underpin a growing number of government services, and the company faces multiple lawsuits alleging that its technology is unsafe (Ars Technica). In this environment, the internal distrust described by The New Yorker could have material consequences for both product development and regulatory compliance. If senior engineers feel unable to raise safety concerns without fear of retaliation or dismissal, the risk of unchecked deployment of powerful AI systems rises sharply. The investigation notes that while no “smoking gun” evidence was uncovered, the aggregate of documented incidents suggests a systemic problem rather than isolated lapses.

Analysts observing the situation point to the juxtaposition of OpenAI’s external messaging and its internal reality as a warning sign. The policy recommendations, while technically sound—calling for transparent risk monitoring, governance frameworks, and public accountability mechanisms—may lack the internal buy‑in necessary for effective implementation (Ars Technica). Without a leadership culture that genuinely embraces those safeguards, the company’s ability to influence AI policy on a global scale could be compromised. As the industry watches, the question emerging from both reports is not merely whether OpenAI can draft responsible AI policies, but whether its CEO can inspire the trust required to enforce them.