Claude Code Flaws Allow Remote Code Execution and API Key Exfiltration

Check Point researchers disclosed multiple critical vulnerabilities in Anthropic's Claude Code that can execute arbitrary shell commands and exfiltrate Anthropic API keys by opening attacker-controlled repositories that include malicious configuration files (hooks, .mcp.json, claude/settings.json and environment variables). The report cites an unnumbered consent-bypass issue and two tracked CVEs (CVE-2025-59536 — RCE; CVE-2026-21852 — info disclosure), explains how exploitation can occur before trust prompts are shown, and notes fixes shipped in subsequent Claude Code releases.