According to a recent LinkedIn post from Cato Networks, the company’s Cato AI Labs is revisiting CurXecute, a vulnerability discovered in 2025 in the Cursor AI code editor. The post describes how a crafted Slack message retrieved through a connected Model Context Protocol (MCP) tool could prompt Cursor to modify its configuration and execute an attacker’s command.
The post suggests that the command could run on a developer’s machine with the developer’s permissions before approval of the change, while noting that Cursor patched the issue following responsible disclosure. It also frames Slack as one example of a broader risk involving MCP servers that process untrusted data, highlighting potential security concerns as AI coding tools become more integrated with external systems.
The disclosure does not provide financial guidance or quantify commercial impact, but it could support Cato Networks’ positioning in AI and cloud-security services. Greater awareness of agentic software vulnerabilities may increase demand for monitoring and protection, although any effect on revenue, customer adoption, or competitive standing remains uncertain based on the post alone.