Claude Code Flaws Allow Remote Code Execution and API Key Exfiltration — Check Point Research Exposes Critical Vulnerabilities

TamilTech AI summary

Check Point Research disclosed two critical flaws in Anthropic’s Claude Code (CVE-2025-59536 and CVE-2026-21852) that can lead to remote code execution and Anthropic API key theft when a developer simply clones a malicious repo and opens it in the tool. The first issue abuses project Hooks and MCP settings so harmful commands can run and trust prompts can be bypassed; the second lets a project override `ANTHROPIC_BASE_URL` so API traffic—and the key in the auth header—goes to an attacker who can still proxy real responses so nothing looks wrong. That matters because it is a quiet supply-chain hit on the developer’s own machine, source, and credentials, not a far-off server breach. Anthropic has shipped fixes, but only people on the latest Claude Code build are protected, so older installs stay exposed. Update Claude Code right away, inspect `.claude/` and `.mcp.json` before you trust a project, rotate your API key if you opened untrusted repos, prefer system-level keys over project config, and watch your Anthropic usage for odd spikes.

• What are CVE-2025-59536 and CVE-2026-21852?
• How can I protect myself from these vulnerabilities?
• Has Anthropic fixed these vulnerabilities?
• Are other AI coding tools affected?

AI-assisted summary, checked by the TamilTech editorial team.