Amazon says agents hunted superintelligence keys, targeted two crypto exchanges


Amazon captured records of credential theft and attempts to bypass identity checks at two cryptocurrency exchanges. One verification provider’s server rejected every recorded deepfake video attempt.
Amazon said records captured by its decoy servers described agents hunting for superintelligence (SI) provider keys and trying to bypass identity checks at two cryptocurrency exchanges. The company assessed with moderate confidence that the agents used ARTEX, a tool designed to test computer security.
The company disclosed the findings on 13 November. Its sensors captured records from 3 October–5 November; the decoys posed as services offering large language models, the software the agents consulted to decide what to do next. Their requests often carried a campaign’s full history, including operator instructions and the agents' reported findings.
Amazon said the records also described activity on systems outside its monitoring. It did not observe that activity as it happened and could not verify the outcomes the agents reported.
An ARTEX operator supplies a target and a goal, and a planning agent divides the work among other agents running security tools. In the credential campaign, the records described stolen keys being used to run models that powered the hunt for more keys, The Hacker News wrote. The agents focused on exposed applications that stored keys for multiple providers, where a weak administrator page could expose many accounts. The records named targets including LiteLLM, One API and Dify.
In the other campaign, the records described agents making deepfake videos and fabricated ID-document photos to try to pass checks at the exchanges. One verification provider’s server rejected the live deepfake video in every recorded attempt. Amazon did not say whether the fabricated ID photos passed document checks.

Attorney General Rob Bonta alleges inadequate safeguards and says a breach of Hugging Face was one of the known incidents. His lawsuit seeks court orders and financial penalties.
A separate FBI alert summarized 11 earlier reports rather than describing a new incident. Its examples included previous attacks on South Korean financial institutions and a compromise involving ConnectWise.
Amazon said the techniques were not new, but that agents changed attackers’ speed and persistence. It said agents succeeded where basic protections were missing, including default credentials, exposed keys, interfaces without authentication and unpatched software.
Amazon said it notified cloud customers who may have been affected. Its team recommended securing SI applications and model access points like other production systems, and watching for unusual usage or spending.
2 min

The government plans to keep the health service’s data-platform contract in place rather than end it immediately, before seeking a new supplier contract.
2 min

Google says attackers used a system called exploit_agent to find and exploit software flaws without human involvement. The server hosting it also held stolen data.
1 min