MONST3R-AI Crimeware Runs Autonomous Pentest Scans on Stolen API Keys, Victims Foot the Bill

What happened

Threat-intelligence firm Bitsight reported on October 15 that a restricted criminal forum is selling an autonomous penetration-testing tool called MONST3R-AI: buyers paste stolen OpenAI, Anthropic, and Google API keys into a control panel, set a budget cap, and provide target lists, while an AI agent autonomously scans the targets for vulnerabilities. The seller hosts no model itself — sidestepping providers' safety training while the scan bills land on the stolen keys' owners. According to Help Net Security's October 16 report, Bitsight tracked thousands of posts across 16 underground forums from February to August 2026.

The seller claims 1,072 findings across 199 scanned hosts. Bitsight states it cannot independently verify that number, but judges the tool effective even if only a small share is accurate — and expects a future trade in stolen keys sold purely to fuel scanners like this one.

The report also charts underground-forum trends: from February to August 2026, mentions of AI agents, indirect prompt injection, and MCP rose 67%, jailbreak talk grew by roughly three-quarters, and discussion of open-weights models like GLM and DeepSeek climbed sharply; ransomware mentions fell 49% over the same window. Bitsight advises treating a leaked AI provider key as seriously as a leaked AWS credential.

Key facts

1. MONST3R-AI: an autonomous penetration-testing tool sold on a restricted criminal forum; buyers paste stolen keys, set a budget cap, and supply target lists while an AI agent autonomously scans for vulnerabilities.
2. Three providers' stolen keys: OpenAI, Anthropic, and Google API keys; the seller hosts no model; scan costs are billed to the key owners.
3. Seller's claimed numbers: 1,072 findings across 199 hosts; Bitsight states it cannot independently verify them.
4. Underground-forum trends: February–August 2026: AI agent/indirect-prompt-injection/MCP mentions +67%; ransomware mentions −49%; open-weights models like GLM and DeepSeek discussion up sharply.