Stolen AI logins flood dark web markets, Google says

Google's Threat Intelligence Group found that underground marketplace prices for stolen Claude, Gemini and Cursor Pro accounts more than doubled in 2026, the Financial Times reported. One vendor, branded "Poison Claude," resells access to Anthropic's Opus and Sonnet models at 5 to 15 percent of the official per-token price.

Poison Claude stocks its accounts by farming free sign-up credits, including AWS Bedrock's $100 welcome offer. It takes payment only in cryptocurrency, according to a summary of the findings published by AI Weekly. Some sellers now offer "guaranteed access," promising a free replacement account if the original one gets banned.

Researchers at Okta said buyers of these resold accounts let the seller "see every prompt and response," and can "modify responses in transit without the customer being able to tell," according to the AI Weekly summary. CrowdStrike's 2026 Threat Hunting Report separately found AI-related adversary activity rose 89 percent between July 2025 and June 2026.

Google also tracked an infostealer called ACRSTEALER targeting API key files from the coding tools Cline and Continue starting in May, according to the report. Security researchers call the practice of running paid models through resold or stolen credentials "LLM-jacking."