AI Hype Under Fire as Experts Challenge Anthropic and OpenAI Claims
Timnit Gebru and Emily M. Bender challenge Anthropic and OpenAI claims on hacking, math breakthroughs and superintelligence, urging independent expert scrutiny.
Summary
On September 22, 2026, DAIR executive director Timnit Gebru and University of Washington linguistics professor Emily M. Bender argued that a summer of AI announcements inflated product capability. At April’s end, Anthropic said Claude Mythos beat most security experts at finding software vulnerabilities. After an OpenAI and Hugging Face hacking incident, Anthropic and Meta disclosed similar cases, often framed as rogue models. Cybersecurity experts instead blamed OpenAI’s failure to use basic safeguards. Anthropic and OpenAI then claimed mathematical breakthroughs. OpenAI said Astra advanced problems whose main result had stalled for at least a decade, but mathematicians disputed its novelty and alleged plagiarism and research misconduct. Two days before a later OpenAI claim, New York University Courant Institute professor Tristan Buckmaster alleged stolen work and improper attribution.
Departing Anthropic engineer Jacob Coxon went viral warning that Anthropic and OpenAI were racing toward self-improving superintelligence and gambling with lives. Gebru and Bender call imminent superintelligence a transhumanist, eugenic fantasy, arguing math and coding attract hype because answers are verifiable and systems can be tuned without paying data annotators. Hundreds of mathematicians warned that commercial incentives encourage exaggerated claims and urged policymakers to consult experts, not press releases. The authors say terms such as rogue model shift agency from companies, obscuring alleged malware creation, academic plagiarism, and customer-data training without consent.
They reject industry claims that bipartisan anti-data-center activism distracts from superintelligence, citing climate damage, nearby asthma, rising electricity bills, public subsidies, and cooling-water use. They call Senator Bernie Sanders’s proposed legislation to prevent artificial superintelligence well-intentioned but misguided, urging slower, independent scrutiny.
Positives
- Hundreds of mathematicians urged policymakers to consult independent experts instead of relying on corporate announcements or popular coverage.
- Cybersecurity experts redirected scrutiny from supposedly rogue models toward OpenAI’s alleged failure to implement established safeguards.
- Anthropic and Meta disclosed model-related security incidents following the OpenAI and Hugging Face hacking case.
- Independent mathematical review challenged initial excitement around Astra and subjected OpenAI’s claimed breakthrough to closer examination.
- Verifiable answers make mathematics and coding useful fields for objectively testing model outputs.
Risks & concerns
- Cybersecurity experts attributed the hacking story to OpenAI’s alleged negligence rather than autonomous model behavior.
- Mathematicians accused OpenAI of plagiarism and research misconduct after finding Astra’s results less novel than initially presented.
- Tristan Buckmaster alleged stolen work and improper attribution two days before a later OpenAI mathematical breakthrough claim.
- Superintelligence framing may shift accountability from companies accused of malware creation, plagiarism, and training on customer data without consent.
- Data centers are linked in the critique to climate damage, nearby asthma, rising electricity bills, public subsidies, and cooling-water consumption.
- Corporate urgency could steer policymakers toward speculative superintelligence threats while sidelining present harms and independent scrutiny.