Evidence · DiggingBeagle record
Amazon Threat Intelligence: ARTEX agent cybercrime campaigns
Amazon MadPot analysis of agent histories across credential discovery and verification-evasion campaigns.
- Published
- Nov 13, 2026
- Source role
- primary disclosure
Evidence record
Amazon MadPot analysis of agent histories across credential discovery and verification-evasion campaigns.
Claim-level citations (4)
- supportsMadPot observed offensive agents seeking inference credentials to fund further automated operations: Amazon collected tool histories and agent-maintained records from attackers interacting with decoy model-compatible APIs. One operation searched exposed AI application, orchestration and provider endpoints for working model credentials. The exposed work plans distinguished candidate secrets, tested working credentials and privileges or available model value. The claimed next step was adding valid keys to a shared inference pool so later reconnaissance and exploitation would consume someone else's compute allowance. This links credential theft to the running cost and persistence of an offensive agent campaign.
Amazon Threat Intelligence, Campaign 1 credential harvesting, key validation and resource pooling
- supportsMadPot observed offensive agents seeking inference credentials to fund further automated operations: Amazon observed offensive tool requests, self-reports and recorded agent context between October 3 and November 5. These are valuable attacker-side artifacts, but they do not independently establish every victim account, billing loss, validated key or ultimate user of the shared pool. The dataset spans different operators and agent roles. Its ARTEX markers are not evidence that a single controller directed every activity or that a named government sponsored the actors.
Primary disclosure: results, scope and limitations
- supportsARTEX agents reproduced face-movement prompts, but recorded KYC checks rejected every synthetic attempt: A separate ARTEX-associated workstream targeted verification at two cryptocurrency exchanges using generated video and forged identification images. Its agent records describe reverse-engineering a liveness vendor's face-motion challenge and testing real-time animation. The best logged synthetic face tracked the expected location within seven pixels; an optimized consumer-GPU pipeline reached approximately 11.5 frames per second. The attackers also manipulated JPEG camera metadata and compression evidence, seeking consistency with photographs rather than screen captures.
Amazon Threat Intelligence, Campaign 2: Defeating identity verification with deepfakes
- supportsARTEX agents reproduced face-movement prompts, but recorded KYC checks rejected every synthetic attempt: The server-side liveness verdict was REJECTED for every attempt shown in the recovered sessions. This is evidence of a coordinated attacker effort and a measured failure of those attempts, not evidence that an exchange authorized a forged identity or released funds. The report draws on an attacker-side decoy interaction. Exchange audit logs, real accounts and any successful cashout were not independently confirmed; the KYC study must not be merged into the separate account-key harvesting operation.
Primary disclosure: results, scope and limitations