Google issued a warning on Monday that cybercriminals are exploiting Large Language Models (LLMs) to fraudulently use valuable computing resources and AI services. "In individual incidents, we observed fraudulent charges associated with resource consumption exceeding $100,000 USD in a single day, with cumulative charges at some organizations reaching $1 million USD before access was revoked." At one end of the spectrum, low-level resource theft involves manipulating customer support chatbots to perform unauthorized tasks, such as code generation.
The most common initial access vector has been exposed or stolen cloud identity tokens, including IAM credentials, Service Account keys, GitHub Personal Access Tokens (PATs), and active session tokens, which allow attackers to establish an initial foothold into target organizations' infrastructure. Run targeted database queries to harvest additional credentials.
A report from Broadcom-owned Symantec and Carbon Black last week reveals a black market service that installs genuine OpenAI Codex and Anthropic Claude CLI tools on enterprise systems, then reconfigures them to route all requests through third-party infrastructure rather than the vendors' servers.
After a legitimate installation of the Codex or Claude Code CLI via npm, users are typically prompted to run a single command to obtain a fileless PowerShell or Bash setup script from a domain resembling "vibe.aiok[.]me." This script overrides the configuration, targeting "~/.codex/config.toml" for Codex or "~/.claude/settings.json" for Claude Code.
