Critics accuse OpenAI and Anthropic of hypocrisy on AI safety

The pattern observed here mirrors historical tensions in dual-use technology sectors where developers simultaneously warn of catastrophic misuse while accelerating deployment. This dynamic previously appeared in gain-of-function virology research and early nuclear energy commercialization, where proponents argued that controlled development was safer than ceding progress to less responsible actors. In the AI domain specifically, this recurs whenever capability jumps coincide with updated safety frameworks. The 'safety-progress paradox' describes the recurring claim that understanding emergent risks requires creating the very systems that generate them. Critics consistently interpret this recursive justification as special pleading, while developers frame it as empirical necessity. This cycle typically intensifies during periods of rapid capability scaling, as the gap between rhetorical caution and observable product velocity widens. Previous iterations involved debates over open-weight releases versus closed APIs, but the fundamental structure—warning of doom while shipping upgrades—remains constant across different technical epochs and organizational leadership changes.

On September 10, 2026, Dr. Tom’s Lens publicly accused OpenAI and Anthropic of hypocrisy regarding their approach to artificial intelligence safety. In a post on X, the critic argued that both organizations appear aligned in predicting catastrophic AI outcomes ('AI dooming') yet continue to release increasingly capable models. Dr. Tom’s Lens characterized this behavior as a 'doom-and-release strategy,' asserting that if these companies genuinely believed AI posed existential dangers, they would cease model improvements rather than continuing development. The critique concluded with a statement of lost respect for both entities and an intention to delete a ChatGPT account held for four years [1]. This specific accusation targets the perceived dissonance between public safety warnings and commercial product velocity.

The criticism from Dr. Tom’s Lens reflects a broader skepticism within the AI community regarding corporate safety narratives. While the primary source text focuses on the ethical inconsistency of releasing dangerous technology, adjacent community discussions highlight parallel concerns about economic motivations overriding safety or quality commitments. For instance, users in relevant forums have drawn comparisons between AI model lifecycles and consumer electronics strategies, suggesting that performance degradation or abstraction may be driven by profit margins and capital expenditure recovery rather than technical necessity or safety alignment [3]. These discussions imply that the 'hypocrisy' accusation extends beyond existential risk to include perceived manipulation of user value and marketing claims.

OpenAI and Anthropic have not issued direct responses to Dr. Tom’s Lens’s specific September 10 post in the provided sources. However, their established defensive positions maintain that continued development is integral to safety research. OpenAI has historically argued that building more capable systems is necessary to understand and mitigate risks effectively, positing that stagnation does not equate to safety. Anthropic defends its approach through Responsible Scaling Policies (RSP), which theoretically allow for progress contingent upon meeting specific safety benchmarks. Critics like Dr. Tom’s Lens reject this framework, viewing it as a justification for continued capability escalation despite acknowledged existential threats.

The controversy also intersects with debates over technical transparency and marketing integrity. Separate community threads reference claims that OpenAI is applying models trained on specific mathematical problems (Navier–Stokes) to unrelated millennium prize challenges like Riemann and P vs NP [2]. Such claims, whether accurate or not, fuel the narrative that safety rhetoric serves as cover for aggressive capability marketing. When combined with accusations of planned obsolescence similar to smartphone manufacturing cycles [3], the charge of hypocrisy gains texture: critics perceive a pattern where safety language masks standard extractive business practices.

Dr. Tom’s Lens’s decision to announce account deletion serves as a symbolic act of protest, signaling that trust has eroded to the point of platform abandonment. This moves the critique from theoretical disagreement to behavioral consequence. The timing suggests a cumulative frustration rather than a reaction to a single new event, as evidenced by the reference to four years of prior usage. The core tension remains unresolved: defenders view iterative scaling as the only viable path to alignment, while critics view it as evidence of bad faith. Without adjudication of whether current safety measures are sufficient, the dispute rests entirely on conflicting interpretations of intent and the validity of voluntary corporate governance in high-stakes technology development.

Dr. Tom’s Lens represents a cohort of four-year veteran users whose public defection threatens the social license underlying voluntary safety frameworks. If such users conclude that 'doom-and-release' is inherent rather than incidental, external regulation becomes the default demand. For OpenAI and Anthropic, the stake is maintaining sufficient trust to avoid prescriptive oversight that could constrain development velocity. The magnitude is currently reputational rather than financial, measured in lost goodwill among technically literate early adopters whose endorsement previously legitimized self-regulation. Continued erosion could shift policy windows toward mandatory pauses or capability caps.