Security experts mark another worrying problem with anthropic AI systems – this is what they have found

• The MCP Inspector project of Anthropic bore an error with which unable to put sensitive data, drop malware
• To abuse it, hackers have to chain with a decades of old browser bug
• The error was resolved in mid -June 2025, but users still have to be wary

The Anthropic Model Context Protocol (MCP) Inspector Project was a vulnerability of critical extraction, which enabled threat actors to set up external code implementation attacks (RCE) against host devices, have warned experts.

Best known for its Claude Conversational AI model, Anthropic developed MCP, a Source Standard that facilitates safe, two-way communication between AI systems and external data sources. It also built Inspector, a separate open source tool with which developers can test MCP servers and debugs.