Ghostjacking attack uses poisoned logs to turn AI agents bad

Demonstrated at DEF CON, Tenet's 'Ghostjacking' attack plants attacker instructions in logs to hijack AI agents, succeeding 9 out of 10 times against Claude Code. It abused Cloudflare, Datadog, and Sentry — hijacking domains, stealing cloud credentials, and turning one AI into an insider that vouched for the attacker.