Prompt injection tags along as GenAI enters daily government use

The report warns that widespread GenAI adoption in government increases exposure to prompt injection attacks—where malicious instructions embedded in content or interactions cause models and agentic systems to exfiltrate data, execute code, or propagate malicious prompts—illustrating the risk with proof-of-concept cases and real incidents (e.g., a VS Code extension issue, GeminiJack, and a worm-style propagation), and recommends policies, least privilege, human approval, and logging to mitigate the threat.