Anthropic sues US over Pentagon AI, xAI restarts, Iran deepfakes

Anthropic filed suit this week against the U.S. government and the Department of Defense challenging a Pentagon artificial intelligence procurement, escalating legal scrutiny of military AI sourcing. In separate developments, Elon Musk's xAI said it is restarting its model effort after multiple internal builds, and Iran-linked actors circulated AI-generated deepfakes targeting political figures.

Anthropic sues over Pentagon procurement

Anthropic says the lawsuit, filed this week, challenges the Pentagon's handling of a contract to acquire advanced language-model capabilities. The company alleges the procurement process raised legal and procedural questions about how models are evaluated and selected for use in defense contexts. Anthropic's filing seeks clarification on procurement rules and how safety and performance were measured in the bid process.

Legal filings name the Department of Defense and related federal agencies as defendants and ask a court to review the award and the underlying evaluations. The case could force federal agencies to disclose more information about model testing, red-team results, and the criteria used to judge competing systems. If the court grants Anthropic relief, contracting timelines and requirements for independent safety assessments could change for future AI purchases.

xAI restarts, Iran-linked fakes spread

xAI announced an internal restart of its model program, acknowledging repeated rebuilds since the company began developing its own large models. The company said earlier iterations were not satisfactory and that engineers will scrap elements of the previous architecture and begin a new development cycle. The restart is the latest in a series of pivots at xAI, which has publicly revised its road map multiple times as it tries to scale model quality and safety.

Observers note the practical costs of repeated rebuilds: lost engineering time, increased compute spend, and delays to product deployments. xAI’s decision follows a period of aggressive hiring and public demonstration that many industry rivals have also faced growing pains while attempting to ship competitive models.

Meanwhile a surge of synthetic audio and video content attributed to Iran-aligned networks circulated on multiple platforms. The material used generative techniques to impersonate politicians and commentators, complicating verification efforts and amplifying misleading narratives. Platform takedowns removed some items, but the rapid spread highlights how readily available generative tools can be weaponized by state and proxy actors.

Security analysts said the content varied in quality, from crude manipulations to convincing clips that could fool casual viewers. The episode prompted renewed calls for stronger provenance metadata, faster verification workflows, and clearer platform policies to limit the reach of harmful synthetic media.

Why it matters

The Anthropic lawsuit raises the prospect of judicial oversight shaping how governments buy and vet advanced AI, which could change compliance and contracting for firms that sell to public agencies. xAI’s repeated restarts illustrate the technical and financial difficulty of building competitive models and signal continued volatility among smaller model providers. The Iran-linked deepfakes episode shows synthetic media remains a potent tool for political influence, increasing pressure on platforms and governments to adopt practical provenance and verification measures.