Gemini Agent-to-Agent Attack Method Exposed Secrets, Enabled Pull Request Tampering

A vulnerability was discovered in the Gemini agent system, where a malicious prompt could be sent to a low-privilege agent, allowing it to pass a harmful comment to a higher-privilege agent. This could potentially expose sensitive information and enable tampering with pull requests. The attack method exploits the interaction between agents with different privilege levels. The vulnerability can be triggered by crafting a specific prompt to a Google ADK agent with low privileges.