Problem
An AI agent could spin up Docker containers with dangerous flags like --privileged, --network host, or volume mounts to sensitive paths.
Proposed Guard
Block or warn when the agent runs docker run with:
--privileged
--network host
-v /:/host or similar root mounts
--pid host
Why It Matters
Container escape via privileged mode is a well-known attack vector. Even well-intentioned agents should not create privileged containers without explicit approval.
Difficulty
Intermediate. Requires parsing docker run flags.
Problem
An AI agent could spin up Docker containers with dangerous flags like
--privileged,--network host, or volume mounts to sensitive paths.Proposed Guard
Block or warn when the agent runs
docker runwith:--privileged--network host-v /:/hostor similar root mounts--pid hostWhy It Matters
Container escape via privileged mode is a well-known attack vector. Even well-intentioned agents should not create privileged containers without explicit approval.
Difficulty
Intermediate. Requires parsing docker run flags.