Connecting five MCP servers to an AI agent raised attack success rate to 78.3%, Unit 42 says
Tests by Unit 42, the research unit of Palo Alto Networks, reported by VentureBeat, showed that connecting five MCP servers to a single AI agent raised the attack success rate to 78.3%. The flaw is not in the model but in the trust chain: MCP standardized how agents discover and call tools, but not how to verify who is calling or what they can touch. A compromised server can inject instructions that travel to later calls, including to servers that were never compromised.
In 45 seconds
- Five MCP servers connected to an agent raised the attack success rate to 78.3% in Unit 42 tests at Palo Alto Networks.
- The mechanism is "cascading trust": a compromised server injects instructions that propagate to later calls, including to servers that were never compromised.
- The MCP specification added OAuth 2.1 authorization in March 2025 and in July 2026 deprecated dynamic client registration in favor of CIMD.
IASummary generated by artificial intelligence from a story by El Ecosistema Startup. It may contain errors: the original source prevails.



Comments
Be respectful. Comments reported by several people are hidden until a moderator reviews them.
You do not need an account to read, only to comment. Sign in with a code sent to your email or with your Google account. We only show the name you choose, never your email.