How AI Coding Tools Crushed the Endpoint Security Fortress

Check Point researcher Oded Vanunu presented findings that AI coding assistants (Claude Code, OpenAI Codex CLI, Cursor, and Gemini) expose a new client-side attack surface by accessing local files and configurations; the team disclosed multiple high-severity vulnerabilities (including CVE-2025-59536, CVE-2025-61260, CVE-2025-54136 and an unassigned Gemini issue) that can enable pre-authorization code execution, RCE, EDR bypass, and plug-in swap attacks. Vendors have patched the reported flaws, and the report urges organizations to audit AI tool usage, sandbox automated shell tasks, and adopt a "Configuration = Code" zero-trust approach for developer endpoints.