Agent session smuggling attack targets AI agent-to-agent communication

A new AI attack, "agent session smuggling," allows rogue AI agents to secretly inject malicious instructions into ongoing communications with victim agents. This technique exploits stateful communication protocols like A2A, enabling attackers to manipulate victim agents over multiple interactions by hiding commands within normal conversation flows. Mitigation strategies include human oversight for critical actions, verifying agent identities, and grounding conversations to detect deviations, as multi-agent systems expand their interoperability.