Пентагон в конце февраля настоял на расширении доступа к моделям у крупных компаний в сфере ИИ, что привело к публичной конфронтации с Anthropic и оперативным действиям OpenAI.

Хронология событий

On 24 February, the Pentagon summoned Dario Amodei (CEO, Anthropic) for a meeting with Hegset (US war minister) and demanded removal of restrictions on Claude.

On 26 February, the Pentagon issued a two-day deadline; Anthropic publicly called the proposed conditions unacceptable and declined to comply.

On 28 February, the deadline elapsed and former president Trump posted an order on Truth Social directing federal agencies to cease cooperation with Anthropic.

Also on 28 February, Sam Altman (CEO, OpenAI) announced an agreement with the Pentagon for access to classified networks under similar operational conditions.

Взаимодействие Пентагона и реакция отрасли

Claude был описан как первый ИИ, допущенный в сети Пентагона, а шестимесячный переходный период указывал на значительную планируемую интеграцию.

On the same day Anthropic was blacklisted, OpenAI closed a private funding round for $ 110B at a valuation of $ 730B, with Amazon investing $ 50B and Nvidia and SoftBank investing $ 30B each.

OpenAI also signed a contract with the Department of Defense while stating prohibitions on autonomous weapons and mass surveillance remain in place.

Проблемы поставщиков и данных

A separate development involved Persona, a KYC provider used by OpenAI, which was reviewed after researchers found code that could transmit data to the US Treasury and query wallets via Chainalysis.

Investigators say the mechanisms discovered in open source could enable data sharing with authorities and vet crypto wallets without explicit user consent.

Последствия и позиции

Anthropic drew a clear boundary on allowed uses and was subsequently designated a "supply chain risk," illustrating the trade-off between access to government contracts and company policy choices.

Some observers note that if US AI firms broadly accept unrestricted military access, that position may strengthen arguments for alternative foreign models elsewhere.

Sam Altman later acknowledged the Pentagon deal was concluded quickly and under conditions he described as rushed and questionable.

Anthropic’s stance prioritized limits on certain military uses, while OpenAI pursued rapid investment and a defense partnership under declared operational constraints.