Researchers found that 1 in 5 MCP access policies came back broken or missing
Bob from finance built a scheduling tool last month. He described it to an AI assistant on a Sunday afternoon, The post Researchers found that 1 in 5 MCP access policies came back broken or missing appeared first on The New Stack .
On July 28, 2026, the Model Context Protocol's maintainers released a specification update focused on authorization, including issuer validation and Client ID Metadata Documents. This update acknowledged that the original trust model was unable to function in a production environment. Bob, a finance employee, built a scheduling tool using the protocol just a few weeks prior without any knowledge of these new authorization measures.
The tool, which relied on default permissions and a previously issued token, became operational across three departments without anyone realizing its potential vulnerabilities. According to recent research, more than 20 percent of the Model Context Protocol (MCP) access policies reviewed were either broken or missing, with many servers using personal tokens rather than service accounts.
These credentials often lacked proper rotation schedules and were not logged, making it difficult to track their usage. Bob's tool, while currently operating without incident, exemplifies the potential risks associated with these unmonitored access policies. The true challenge lies in determining which AI-built integrations pose a genuine threat, rather than prohibiting them altogether.
Written by urgent.news from The New Stack's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
