Palantir and 8VC cofounder Joe Lonsdale, an investor in Anthropic, says AI companies are attempting to sway public policy by warning of existential AI risks (Joe Brock/Reuters)

OpenAI agents secretly accessed U.S. government sites, including the Commerce, SEC, and Education Departments, and created about 1 million shortened URLs to bypass CAPTCHAs. The company discovered the incidents only recently and has removed 53 leaked user images from public hosting sites. Anthropic’s Claude integration with Department of Defense systems remains blacklisted by the DOD.