Skip to content
SurenoGet Sureno Browser
← Sureno News

Attackers using AI agents automate cybercrime at industrial scale, FBI and AWS warn

Agents harvest access keys

In one campaign, attackers used AI agents to steal API keys and pass them to ARTEX’s model gateway. That access enabled further agents to search for additional keys to AI models. AWS Chief Information Security Officer CJ Moses said the agents checked identities through provider services, examined storage and permissions, and pursued managed AI services, secrets, parameter stores and application identities.

Deepfakes fail identity checks

In a separate campaign, attackers used ARTEX to produce real-time deepfake videos aimed at KYC checks at two cryptocurrency exchanges. They also forged ID photos to try to pass document verification.

The agents completed the movement sequence with the face within 7 pixels of its target, but the exchange vendor’s server-side checks rejected every deepfake attempt. For document checks, agents created synthetic camera metadata matching the JPEG’s actual image dimensions and retained original camera compression tables, which edited images often lose when saved again.

Campaigns and exposed consoles

That campaign ultimately compromised more than 2,250 downstream Microsoft tenants. The article says the techniques themselves were not new, but that AI agents can make attacks faster and more persistent.

Forked and localized versions of ARTEX, as well as a commercial version, remain available. On November 4, Censys identified 1,465 hosts with an exposed ARTEX console; the FBI said the total is likely significantly higher when private and firewalled systems are included.

Recommended defenses

The guidance calls for reducing external attack surfaces by limiting sensitive services to trusted networks, private connections or dedicated administrative paths. It also recommends placing enterprise applications behind zero-trust access policies and permissions.

Key points

  • FBI and Amazon Threat Intelligence described attackers using ARTEX to automate campaigns against critical sectors.
  • Agents stole API keys and sought access to more AI models and services.
  • Deepfakes failed KYC checks at two cryptocurrency exchanges, despite agents’ attempts to refine them.
  • Censys found 1,465 exposed ARTEX consoles on November 4.
Read the original on theregister ↗

More news

Loading stories…