Conspiracy theories and speculation rife about Anthropic's policy strategy
Update (26 June 2026). The US government has partially reversed its export-control suspension, notifying AnthropicDealroom has a profile for this one. Try Dealroom → that its strongest cybersecurity model, Mythos 5, can be redeployed to a set of "trusted" US organisations that operate and defend critical infrastructure. A source familiar with the directive said more than 100 companies and institutions — many of them Fortune 500 firms, and a number tied to Anthropic's "Project Glasswing" cohort of roughly 100 well-known tech companies — will regain access; Fable 5 is expected to return for general use later, on an unclear timeline. In a letter, Commerce Secretary Howard Lutnick cited "significant progress" on the model's risks and said an export licence would no longer be required for Mythos 5 to reach trusted companies and their non-US-citizen staff, while restrictions remain for firms off the approved list. The government's role in choosing which companies gain access drew criticism: OpenAI CEO Sam Altman wrote that extensive safety testing "is not a bad idea" but that he disliked "the idea of the government picking the customers", and the Foundation for Individual Rights and Expression questioned the lack of transparency and "rule of law". Separately, OpenAI said it was delaying a full public launch of GPT-5.6 at the government's request, limiting it to a small group of vetted partners. The move lends concrete weight to the speculation rounded up below about how export controls are shaping the competitive landscape.
As Anthropic sits at the centre of the US AI export-control debate — following the withdrawal of its Fable 5 and Mythos 5 models under federal restrictions in 2026 — a number of high-profile investors and commentators have publicly speculated about the company's lobbying and competitive strategy. The following is a roundup of those views; they are opinion and conjecture, not confirmed facts about Anthropic's conduct.
Bill Gurley's 'regulatory capture' critique. Speaking on the All-In Podcast in May 2026, Benchmark partner Bill Gurley called Anthropic "one of the most aggressive lobbying startups of all time", arguing it is pursuing regulatory capture to entrench its position against cheaper rivals. He claimed the firm is leaning on federal export controls rather than the courts to address model distillation by competitors.
The 'Dr. Frankenstein theory'. After roughly 30 days reviewing Anthropic's public materials — including CEO Dario Amodei's essay "Machines of Loving Grace" — Gurley said he believes some of the company's leaders are "excited about building a species that's superior to humans", framing the work as "midwifing a deity" rather than writing software. Co-panellist Chamath Palihapitiya separately likened advanced AI to a "digital super-god".
The 'sleeper circuits' thesis. AI commentator Teortaxes (@teortaxesTex) argued in June 2026 that the US government could permanently withhold its most capable "Mythos class" models from general release and still neutralise Chinese competition — by casting any leaked or fast-follower copies as compromised with "CCP sleeper circuits" and unknown cyberattack potential, thereby making them untrustworthy to deploy. He illustrated the point with the 2024 "Sleeper Agents" research paper on deceptive language models, suggesting such a narrative could "destroy the market for frontier Chinese models, even inside China".
These commentaries reflect an ongoing public debate about whether AI-safety advocacy and national-security framing also serve the commercial interests of leading US labs. Anthropic has not issued an official response to the regulatory-capture characterisations.
Read more: Semafor · Reuters · Teortaxes on X · Benzinga / All-In Podcast · Fortune
