Apex npm tools pair malware with AI traffic hijacking
Elastic Security Labs tied two Apex-branded npm packages to REF8801, a campaign delivering information stealers and backdoors to Windows and macOS while routing AI-agent requests through attacker-run services. The packages, @apexacc/cli and @apexacc/audit, were advertised as a Web3 coding assistant and an 18-tool security scanner.
The fake assistant offered access to Anthropic Claude and Google Gemini but sent model requests through services controlled by the operators, putting prompts in their path. The audit package also returned fabricated security findings, so teams could mistake invented results for evidence about tokens, wallets, or contracts.
For teams that ran these packages, removing them would not establish what prompts already passed through those services or whether decisions based on the audits were sound. Organizations evaluating npm-distributed coding agents or Model Context Protocol (MCP) servers face the same trust boundary: a package can mediate both executable code and sensitive model traffic.