High‑Severity Claude Code Desktop Flaw Enables Malicious HTML to Harvest Developer Credentials
What Happened — Researchers at Noma Labs disclosed a high‑severity vulnerability in Anthropic’s Claude Code Desktop application. A crafted HTML file, when opened locally, can execute JavaScript that reads stored developer credentials (API keys, tokens) from the client’s environment and exfiltrate them to an attacker‑controlled server.
Why It Matters for Trust & Control Assurance —
- The flaw illustrates a gap in the protect function of continuous control‑assurance programs: untrusted local content can bypass application‑level access controls and expose privileged secrets.
- Demonstrates the need for continuous monitoring of endpoint security controls and evidence that credential storage follows least‑privilege and isolation principles.
- Aligns with the Access Control control objective in the Verisq Common Framework, which maps to the Identify/Protect functions of NIST CSF 2.0.
Who Is Affected — SaaS and AI‑focused technology firms, development teams using Claude Code, and any organization that stores API keys or tokens on developer workstations.
Recommended Actions —
- Immediately apply any patches or mitigations released by Anthropic.
- Enforce strict file‑type validation and sandboxing for local HTML rendering.
- Review and tighten credential storage policies; ensure secrets are kept in vetted secret‑management solutions rather than local caches.
- Capture evidence of remediation steps for audit readiness. Source: HackRead
Technical Notes — The vulnerability stems from an insecure handling of the file:// protocol within the desktop client, allowing arbitrary JavaScript execution. No CVE has been assigned yet; the issue is classified as high severity by the researchers. Affected data includes developer API keys, OAuth tokens, and other locally stored secrets. Source: HackRead