Wiz says AI agents are running hands-on intrusions
Original title: The Rise of Agentic Attacks: What They Look Like, and Where They're Headed
Named cases include ARTEX at Korean banks and ConnectWise, Hermes at Thailand's finance ministry, and AI-assisted API attacks in Japan.
Wiz says incident response and threat hunting increasingly show AI agents doing hands-on intrusion work under human objectives, from exposed services through cloud and Kubernetes credential theft to bulk secret theft. Public cases include ARTEX against South Korean financial institutions and, per Microsoft, Storm-0940's ConnectWise Cloud Backup supply-chain activity, plus Hermes logs in an intrusion at Thailand's Ministry of Finance. JPCERT/CC, CrowdStrike and DIVD describe further AI-assisted or suspected agentic attacks. Wiz says the activity is often noisy and hard to separate from legitimate automation.
Unconfirmed: Autonomy and evidence strength vary, and some cases do not identify the agent or framework.
- Korean cases reportedly exposed data on about 67,000 people, according to cited researchers.
- Microsoft said ARTEX activity tried sign-ins against more than 2,250 tenants.
- Privileged cloud access was achieved in under three hours in multiple cases, Microsoft reported.
- Hermes ran unattended; initial access was unknown and no exfiltration was found then.
- CrowdStrike described four further actors using agents against healthcare, government and private firms.
Impact: Credential theft, cloud access and data or secret exfiltration
- 1.Correlate noisy automation-like activity into a single intrusion picture.
- ATT&CK
- T1110.003 Password SprayingT1190 Exploit Public-Facing ApplicationT1505.003 Web ShellT1046 Network Service Discovery
- Vendors
- ConnectWiseMicrosoft
- Threat actors
- Storm-0940
- Victims
- Thailand Ministry of Finance
- Organizations
- WizCrowdStrikeJPCERT/CCDIVDHunt.io
From the original article
13 min read at wiz.io
AI agents are changing how attackers operate. Explore real-world cases of agentic attacks, learn how to identify their fingerprints, and discover what they mean for the future of cybersecurity.
By Avigayil Mechtinger · wiz.io
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.