A new exploit called “Friendly Fire” hijacks Anthropic’s Claude Code CLI and OpenAI’s Codex CLI to execute attacker-controlled binaries during routine security reviews of third-party code. The attack ...