Security

Widespread Flaws Let Prompt Injection Hijack AI Coding Tools and CI/CD Pipelines

New disclosures show agentic models can mistake repository content for commands, enabling high‑impact exploits.