Anthropic acted without bad motive, and lost 2-1.

The court held that withholding Claude features can qualify as a supply chain risk.

The panel applied section 4713, which covers any person, not section 3252, which a California judge found requires adversary malice.

A California federal judge ruled last month the parallel designation was illegal because Anthropic is not a supply chain risk.

Anthropic is weighing rehearing by the full appeals court or a petition to the US Supreme Court.

How each outlet framed it
Ars Technica
frames AI trade-off: unconstrained models risk hallucinating lethal targets vs constrained models fail critical military operations

Sources: Ars Technica, Hacker News