Widely available open-source artificial intelligence tools are enabling attackers to search corporate networks for weaknesses around the clock, without constant human supervision. AI-assisted hacking is still in its early stages, but free-to-download tools could already change the cost and speed of cybercrime.
Attackers can use agents that independently adjust attack plans, seek vulnerabilities and steal credentials. They may do so before human defenders realize their systems are being probed.
Rapid intrusions and stolen access
The Federal Bureau of Investigation (FBI) warned on Friday that hackers have used an AI framework called Artex to attack organizations worldwide. In some cases, attackers moved from selecting a target to obtaining privileged access in less than three hours.
In a separate report on Friday, Amazon said attackers used Artex to steal credentials for AI models, then used the resources they obtained to power further attacks. Developer platform Railway and blockchain infrastructure provider Chainstack also reported attacks involving Artex last week.
Palo Alto Networks said on Thursday that Chinese-speaking hackers used another open-source framework, Cairn, to conduct 25 operations against at least 19 organizations between August and September. In a separate report on Monday, Google said Cairn had also been used during an intrusion into the U.S. aerospace sector.
CrowdStrike observed another group using Artex to target more than 200 organizations worldwide over 11 days last month.
How the agents work
Artex and Cairn use AI agents to automate tasks that attackers previously performed themselves. A user supplies a target and an objective, such as entering a corporate network. The agents then develop an attack plan, search for vulnerabilities and use hacking tools to test possible entry points.
The agents share their findings and adjust tactics when an approach fails. This allows them to operate around the clock without constant human supervision. Amazon's findings indicate that attackers can also use stolen access to AI models to search for more credentials, expanding their operations without paying for computing capacity.
Sam Rubin, senior vice president of Palo Alto Networks' Unit 42 threat intelligence and consulting unit, told Axios that the company is seeing more active attacks than ever. He said open-weight models are approaching the capabilities of the most advanced models and, when combined with tools such as Artex or Cairn, are placing a heavy burden on organizations.
John Hultquist, chief analyst at Google's threat intelligence group, told Axios that incident response firms worldwide are operating at capacity, with some forced to turn away cases because they lack resources.
Not every incident involves AI
Digital services remain vulnerable even when there is no evidence of an AI-assisted attack. Over the past week, organizations reporting outages have included Microsoft, Apple, Chase, United, Workday, the Federal Aviation Administration and the National Security Agency.
According to Report URI, an unknown group used the Optimizely Web experimentation platform to infect approximately 178 major websites with cryptocurrency-stealing malware in less than a day. Affected sites included those of Zoom, the BBC, CNN, Salesforce, Gitlab and Ikea. There is currently no indication that AI was involved in that incident.
Speed may matter more than concealment
Hultquist said autonomous attacks generate substantial suspicious activity, potentially making them easier to detect. Financially motivated attackers, however, may only need to steal valuable information before defenders intervene. He also warned that state-backed hackers could use these capabilities to disrupt critical systems before security teams can respond.
The pace of automated attacks may require security teams to fundamentally rethink their defenses.



