FBI and AWS Warn That Attackers Are Automating Cyberattacks Using AI Agents

Attackers are using artificial intelligence tools to run cyberattacks quickly and on a large scale against banks, government agencies, and healthcare organizations around the world.
Security alerts issued by the Federal Bureau of Investigation and Amazon Threat Intelligence show that malicious groups are relying on an open-source framework known as ARTEX. Between October 3 and November 5, threat sensors run by Amazon detected two separate campaigns using this framework to steal credentials and bypass security checks.
In the first campaign, digital agents searched for and stole digital access keys, feeding them back into the framework to locate and compromise more managed AI services and application identities. AWS Chief Information Security Officer CJ Moses described the scope of the operation in a blog post, explaining that the automated tools checked provider tokens, reviewed storage permissions, and targeted application credentials.
Targets included language model gateways, proxy platforms, and popular AI application builders. The FBI noted that the framework speeds up attacks significantly, allowing some intruders to move from initial targeting to privileged system access in under three hours. Recent examples include August attacks against South Korean financial institutions that compromised personal data for roughly 67,000 people, as well as a September supply chain attack targeting backup customers.
In the second campaign, attackers used the framework to generate real-time deepfake videos and forged identification documents in an effort to trick identity verification checks at cryptocurrency exchanges. According to Moses, the automated agents reverse-engineered vendor liveness protocols to understand face-detection event streams and coordinate systems, though server-side verdicts ultimately rejected the deepfake attempts.
Although the original software developer removed its repository in October, forked variations, localized versions, and a commercial edition continue to circulate. Internet scanning services recently identified more than 1,400 exposed consoles running the software, with actual deployments likely much higher behind private networks.
Federal authorities urge organizations to reduce their external attack surfaces by restricting sensitive services and moving enterprise applications behind strict access policies. Moses emphasized that the automated threats succeeded primarily where basic security practices were overlooked. "The agents succeeded where basic security hygiene was missing, including default or hardcoded credentials, keys left in code and public artifacts, unauthenticated interfaces and inference endpoints, and unpatched software," Moses added. "Getting those fundamentals right remains the most effective defense".
Summary
- Automated AI agents allow attackers to chain cyberattack techniques much faster, moving from initial targeting to deep system access in hours.
- The underlying security vulnerabilities exploited by these campaigns involve basic hygiene flaws like default credentials, unpatched software, and exposed endpoints.
- Despite the removal of the original open-source repository, forked variations and exposed consoles remain active across the internet.