Researcher Finds Cross-Agent MCP Trust Flaw Across Google, JPMorgan Chase and Government Servers

An independent security researcher says multiple AI-agent deployments have confirmed a shared weakness in how Model Context Protocol servers trust agent-supplied data, allowing one compromised agent workflow to hand malicious instructions or unsafe requests to another system that treats them as internal. The findings span fixed SSRF bugs at Google, JPMorgan Chase, Weaviate, France's interministerial digital directorate and Tangerang City government, plus a separate Rapid7 CVE and five unresolved U.S. federal findings still in triage.