Impersonation Site Targets Claude AI Users with Credential‑Stealing Malware
What Happened – A malicious website masquerading as Anthropic’s Claude AI chatbot was observed distributing a credential‑stealing payload (referred to as an “ACR stealer”). The site lures visitors with a fake Claude interface, captures login credentials, and may install additional malware.
Why It Matters for TPRM –
- Third‑party AI services are increasingly embedded in business workflows; compromised credentials can expose sensitive corporate data.
- Credential theft on a vendor‑facing portal can lead to lateral movement into client environments.
- The attack demonstrates the risk of supply‑chain exposure through seemingly benign SaaS tools.
Who Is Affected – SaaS providers offering AI/ML APIs (e.g., Anthropic Claude), their enterprise customers, and any organization that integrates Claude into internal applications.
Recommended Actions –
- Verify that all Claude API keys and user credentials are stored in a password‑manager or vault with MFA.
- Conduct phishing‑simulation training focused on AI‑tool impersonation.
- Review and tighten web‑traffic filtering for domains resembling “claude.ai” or related brand names.
Technical Notes – The threat leverages a phishing‑style impersonation page (likely hosted on a compromised or newly registered domain) to deliver a JavaScript‑based credential harvester. No specific CVE is cited; the vector is social engineering rather than a software flaw. Data types at risk include API keys, user passwords, and potentially uploaded prompts containing proprietary information. Source: SANS Internet Storm Center