←Home

    Guardrails for AI Agents

    How researchers attack, monitor and constrain autonomous LLM agents that click, type and call tools in real environments, from prompt injection and quiet data leaks to runtime monitors and enforceable safety policies.

    In order

    1 stories
    small-plane cockpit: attitude indicator, compass and gauges against clouds
    small-plane cockpit: attitude indicator, compass and gauges against clouds

    Small-plane cockpit: attitude indicator, compass and gauges against clouds · Photo: Rinald Rolle

    Alignment · Agent prompt injection

    Office AI agents run planted scripts when users ask, a shared-workspace benchmark finds

    A 2026 test of 12 AI models doing finance, legal and consulting work found that swapped or attacker-supplied scripts derailed most of them.

    arXiv preprint · 9 October 2026