Anthropic

483 articles in 2026 Q2 marked Anthropic’s shift from an AI-model maker into a focal point of US defense, safety, export-control and compute-capacity disputes.

Who they are

Anthropic is an AI company whose coverage centers on its Claude products and newer Fable and Mythos models, with Dario Amodei serving as its prominent executive voice. The company appears both as a leading model developer competing with OpenAI and as a consequential supplier whose safety policies have drawn scrutiny from US defense and national-security institutions.

The recent arc

Coverage accelerated sharply in 2026, rising from 332 articles in Q1 to its 483-article peak in Q2, as a dispute with the Pentagon put Anthropic’s deployment rules at the center of the story. Amodei said the company could not “in good conscience” remove safeguards sought by the DOD, while Defense Secretary Pete Hegseth directed a supply-chain-risk designation. Reporting on the failed talks said the Pentagon sought to use Anthropic AI to analyze bulk data collected about Americans; OpenAI then contrasted its own DOD arrangement with Anthropic’s guardrails.

The recent arc

The story then moved from a defense-access conflict to a broader test of model controls and infrastructure. Anthropic debuted Claude Fable 5 and restricted Mythos 5 to trusted organizations, described fallback safety classifiers for sensitive areas such as cybersecurity, and announced Project Glasswing. Days later, it said it was disabling Fable 5 and Mythos 5 after a US export-control order, amid reports of White House talks to resolve the Mythos dispute. By late August, attention had also turned to expansion: Bloomberg reported a rumored $45 billion, six-year Nscale capacity commitment in West Virginia, while reports tied a prospective IPO pitch to claimed revenue opportunities above $30 trillion.

The tension

The central tension is whether Anthropic can preserve its stated safety boundaries while remaining viable in the government and frontier-model race. Its conflict with the Pentagon and Trump administration made model access, cybersecurity safeguards and data use matters of public policy, while OpenAI became the clearest competitive counterpoint through its own DOD agreement. The prospective Nscale commitment and work toward custom semiconductors add a second contest: securing enough compute to compete at the frontier without making infrastructure dependence and regulatory exposure still more consequential.

Why it matters

If this trajectory holds, Anthropic’s product decisions will increasingly be shaped not only by enterprise adoption but by the terms under which governments permit, procure or restrict advanced AI. The Fable and Mythos episode shows how quickly safety claims, alleged misuse and export controls can affect availability, while the scale of reported compute plans underscores that leadership may depend on power and chips as much as model capabilities. Whether Anthropic can resolve those pressures without yielding the safeguards it defended remains the key uncertainty.

Related: AI · Claude · OpenAI · Pentagon · Dario Amodei says Anthropic cannot “in good conscience” accede to DOD' · Anthropic says it is disabling Fable 5 and Mythos 5 for all customers