Google says threat groups now automate attacks with AI pentest agents
Original title: Google warns that fully-automated AI attacks are coming soon
GTIG says exploit_agent ran without a human and projects low-interaction attacks within three months.
Google Threat Intelligence Group says criminal and state-sponsored groups now use open-source multi-agent pentest suites, including CAIRN, ARTEX, CyberStrikeAI, and HexStrike AI, against low-complexity flaws on exposed sites. GTIG found an attacker staging server for exploit_agent that automated research and exploitation of French media, entertainment, and education sites and stored stolen data and webshells. The FBI said actors have weaponized ARTEX since August across several sectors. UNC6836, with suspected ties to Midnight Neptune, changed Claude Code so launch downloaded the SOMBERMEME backdoor at crypto firms.
Unconfirmed: North Korean involvement and UNC6836's link to Midnight Neptune are suspected; the timelines are GTIG estimates.
- GTIG estimates little-to-no-human attacks within three months and autonomous ones within six.
- CAIRN was used in an attack on the U.S. aerospace sector.
- Suspected North Korean hackers used Claude Code to deliver malware.
- Hundreds of mostly AI-generated PHP2Shell proof-of-concept exploits followed PHP's flaw announcement.
- Google also reports AI in influence operations and widespread LLMjacking.
Impact: Automated exploitation of exposed web apps and backdoor delivery
- 1.Treat purported PHP2Shell proof-of-concept code as unverified until checked.
- Threat actors
- UNC6836Midnight Neptune
- Malware
- SOMBERMEME
- Organizations
- Google Threat Intelligence GroupFBI
- Countries
- United StatesNorth KoreaFrance
From the original article
3 min read at bleepingcomputer.com
Threat actors are now relying on agentic workflows powered by AI-based offensive frameworks to automate vulnerability discovery and exploitation, accelerating attacks to machine speed. [...]
By Bill Toulas · bleepingcomputer.com
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.