MCP marketplaces shipped LOLBAS malware. We audited 256 agents.

A security audit found that a public MCP marketplace shipped an agent using a LOLBAS chain to fetch and run remote payloads via a signed Microsoft binary.

Because the payload stayed hidden inside trusted system binaries, normal static scans could miss it, creating a supply-chain and endpoint-security risk.

Researchers audited 256 agents with 27 checks and positioned Trust Agent as a vetted alternative.

The case highlights how AI agent marketplaces can become a delivery channel for malware-like behavior and signed binary proxy execution.

It maps closely to MITRE ATT&CK T1218 and raises the bar for marketplace security and agent vetting.