AI Wiki · Mediocre Security
IncidentIncidents

ARTEX campaigns seen by Amazon MadPot (Oct-Nov 2026)

Amazon's decoy sensors captured ARTEX agents' own campaign records: industrial-scale theft of AI provider keys that paid for further hunting, and a deepfake attack on crypto-exchange identity checks.

What happened

From Oct 3 to Nov 5 2026, Amazon’s MadPot network of decoy sensors, set up to look like popular LLM inference APIs, received requests from AI agents built on ARTEX. Because agent frameworks send their whole working context to the model, each request carried the agents’ system prompts, tool histories, shared knowledge graph and reported findings, in effect the campaign’s own log. Amazon published what those records describe on Nov 13 2026, the same day the FBI issued a flash alert on ARTEX. Amazon stresses that it did not watch the activity happen and cannot verify the outcomes the agents reported.

Amazon assesses with moderate confidence that the operators were separate Chinese-nexus groups, citing simplified-Chinese prompts, China time-zone settings and targeting rules that excluded mainland China and government domains.

Campaign 1: stealing AI keys at scale

The largest campaign stole API keys for AI models, likely so operators could use frontier models at someone else’s expense, a form of LLMJacking. Stolen keys went into the framework’s own model gateway and powered the agents searching for more. The agents looked for internet-exposed AI applications that hold upstream provider keys, especially gateways and proxies that aggregate many keys (One API, New API, LiteLLM and others) and app builders such as Dify, Langflow, RAGFlow, Open WebUI and Ollama. Their playbook mapped each product to known attack paths, including unclaimed fresh installs, default admin credentials such as LiteLLM’s default master key, CVE-2025-3248 (Langflow), CVE-2026-28797 (RAGFlow), a header-based auth bypass in One API-family systems, SSRF to cloud metadata, and secrets left in client-side code. Discovery drew on internet scanners, certificate transparency, GitHub, package registries, container images and paste sites. One operation graded each credential by whether it worked and what it could reach, and told agents to treat keys public for over 30 days as low value.

Campaign 2: deepfakes against KYC

A second campaign targeted identity checks at two cryptocurrency exchanges. Agents reverse-engineered one vendor’s liveness protocol, built a real-time deepfake video pipeline on a single consumer GPU (11.5 frames per second, per the records) and forged camera metadata on edited ID photos. At least four agent sessions ran in parallel and handed off work. The vendor’s server rejected the deepfake in every attempt the agents recorded.

Why it matters

Amazon says none of the techniques were new; what changed was speed and persistence, with operations running for weeks with little human input. Its advice: assume exposed secrets will be found quickly, and bring AI gateways and inference endpoints under the same controls as other production systems. The two CVEs are listed in the AI CVE tracker.

Timeline

Every dated event in the wiki that involves ARTEX campaigns seen by Amazon MadPot (Oct-Nov 2026), newest first.

Connections

ARTEX campaigns seen by Amazon MadPot (Oct-Nov 2026) and everything one link away. Drag to move, scroll to zoom, click a node to open it.Open in the full graph →

What links here

Concept AI and CVEs: latent flaws, now exploitable

Company Amazon (AWS)

Product ARTEX

Government body FBI

Incident LiteLLM guardrail sandbox escape (CVE-2026-40217)

Sources

  1. AWS Security Blog, “Amazon Threat Intelligence observes threat actors running cybercrime campaigns with ARTEX AI agents”, Nov 13 2026
  2. The Record, “FBI flash alert says Artex tool automating cyberattacks at 'machine speed'”, Nov 13 2026