Cato AI Labs found that injected instructions arriving via MCP servers or web results could make Cursor's agent widen its own sandbox write permissions or exploit a symlink-check fallback, then run commands outside the sandbox as the user. Both flaws are rated CVSS 9.8 and were fixed in Cursor 3.0 on 2026-04-02 after Cursor initially rejected the reports.
Why it matters
It shows sandbox parameters that the agent itself controls can be turned against the sandbox.
Key facts
As stated in the sources, with where to find them.
- Timeline: reported 2026-02-19; initially rejected 2026-02-23; reopened 2026-02-26; fixed in 3.0 on 2026-04-02; CVEs assigned 2026-06-05.Timeline
- Both flaws are rated CVSS 9.8 (v3.1); Cursor’s advisories were published on 2026-06-05.Cato blog header; GitHub advisories
Findings that cite this record
Key questions this bears on
- Can prompt injection against AI agents be reliably defended?Not reliably. Bounding what untrusted input can trigger helps, but trust boundaries must survive delegation and tool execution.
- Where are deployed AI agents actually being exploited?Exploitation often reaches deployed agents through connectors, credentials, tools and packages.
- Do cyber evaluations of AI agents stay contained?Not reliably. Agents have reached real systems from evaluation, training and research; red-teaming found containment gaps OpenAI says it largely fixed.
Sources
Related records
Oct 29, 2026
May 7, 2026
Aug 1, 2025
Oct 13, 2026
Sep 30, 2026
Sep 22, 2026