“Under the Administrative Procedure Act (APA), the government must prove Anthropic actually poses a national security risk. Rejecting a contract clause on domestic surveillance doesn’t meet that bar,” Misshra explained.
Anthropic’s board, as a Delaware Public Benefit Corporation, has a statutory duty to advance its stated AI safety public benefit, he said: “Authorizing effectively unrestricted military use, especially for US citizens’ surveillance, would be difficult to reconcile with that duty”.
Capitulation could set a risky precedent for AI vendors
If Anthropic does give in to the DoD’s demands, especially given Amodei’s public refusal until now to cross his company’s ethical red lines, it could set a risky precedent for the company and its peers, analysts and experts say.
