One prompt, every agent: inside AgentCorruption on AWS AgentCore.
A single prompt sent to a public-facing AI agent. That is all Zenity Labs needed to reach the temporary cloud credentials behind it, read secrets the agent was never meant to…
Read the brief