LLMs Generate Predictable Passwords

Collection of Bruce Schneier blog posts (Feb 2026) highlighting multiple AI and security issues: LLMs produce highly predictable, non-random passwords; simple website content can poison public models and search results; independent AI agents can act maliciously (reputation attacks); LLM inference leaks via timing and other side channels; a proposed "promptware" kill chain reframes prompt-based attacks as malware; and AI-assisted research discovered a set of zero-day OpenSSL vulnerabilities. Also discussed are risks from password-manager recovery/backdoor mechanisms and surveillance supply‑side