Supply Chain Attack on LiteLLM and Anthropic Source‑Code Leak Highlights Enterprise AI Security Gaps

Supply Chain Attack on LiteLLM and Anthropic Source‑Code Leak Highlights Enterprise AI Security Gaps

What Happened – In April 2026 two high‑profile AI incidents were disclosed: a packaging error at Anthropic leaked internal files and source code for its Claude Code assistant, and a malicious commit to the open‑source LiteLLM library (used by many AI‑enabled applications) injected credential‑stealing malware, exposing customer data flowing through AI services.

Why It Matters for TPRM –

• Human‑error releases and open‑source supply‑chain compromises can give attackers deep insight into proprietary AI models and the data they process.
• Third‑party AI components (e.g., LiteLLM) are now a critical attack surface for enterprises that rely on them for internal workflows.
• Loss of source code and API‑key theft can lead to long‑term espionage, model‑theft, and credential abuse across multiple vendors.

Who Is Affected – Technology / SaaS providers that embed AI models, API‑provider ecosystems, and any enterprise that integrates AI‑driven applications (finance, healthcare, retail, etc.).

Recommended Actions –

• Conduct an inventory of all AI‑related third‑party libraries and enforce strict version‑control and code‑signing policies.
• Review vendor security‑by‑design practices, especially around release packaging and open‑source dependency management.
• Implement continuous monitoring for anomalous API‑key usage and enforce least‑privilege access for AI service credentials.

Technical Notes –

• Anthropic leak: Human error during release packaging caused publicly accessible internal files and source‑code exposure. No known CVE; risk stems from lack of secure build pipelines.
• Mercor attack: Malicious code inserted into the LiteLLM open‑source repository (vulnerability‑type supply‑chain exploit) harvested API keys and redirected data to attacker‑controlled endpoints.
• Data types exposed include proprietary model code, configuration files, and customer data processed by AI services.

Source: Proofpoint Threat Insight – Anthropic Leak and Mercor AI Attack