AI-enabled attacks

Cisco Talos researchers describe CLOSEDQUORUM, a malware binary they call the first "LLM-as-C2" architecture, which uses a panel of large language model judges to automate command-and-control and steal credentials. It targets LSASS credential dumping and browser saved passwords in Chrome, Edge and Firefox, plus crypto wallets such as MetaMask, Ethereum and Exodus. Cisco Talos states there is not yet any confirmation of its deployment in the wild.

A likely Russian-speaking threat actor used hundreds of AI agents, built on OpenAI's Codex and DeepSeek models with commodity offensive tools, to exploit CVE-2026-81578 and CVE-2026-82078 in PaperCut NG/MF servers starting August 31, according to GreyNoise. GreyNoise says the campaign compromised at least 440 PaperCut instances tied to 395 organizations in 48 countries, with administrator privileges obtained at 12 organizations, mostly in education.

Dream, a cybersecurity firm, reported that multiple AI agents built on Hermes and OpenClaw ran a near-autonomous intrusion campaign against government networks in Asia over four days in early July. The agents produced 1,395 files, cracked 85 credentials, and exfiltrated thousands of personnel records. Taiwan's Ministry of Digital Affairs separately reported an AI agent-assisted attack on government agencies in the same period, though neither party has confirmed a link between the two.

Unit 42 identified an AI-enabled autonomous hacking campaign by a Chinese-speaking threat actor, using the aliases knaithe and KnYuan. The actor used DeepSeek through the Hermes Agent framework, controlled via Telegram, to enumerate targets with FOFA, download exploit code and attempt exploits against infrastructure using seven vulnerabilities. When initial exploitation failed, the agent autonomously searched for critical-severity CVEs and pivoted to higher-value targets.

A newly identified open source platform, CyberStrikeAI, packages end-to-end attack automation into an AI-native orchestration engine with over 100 curated tools covering the whole kill chain. Team Cymru links its developer, known as Ed1s0nZ, to the threat actor behind a campaign that breached hundreds of Fortinet FortiGate firewalls, and says the developer has some ties to the Chinese government. The researchers observed 21 unique IP addresses running CyberStrikeAI between January 20 and 26.

Team Cymru reports that CyberStrikeAI, a newly identified open-source AI security testing platform, was run by the threat actor behind a campaign that compromised more than 500 Fortinet FortiGate firewalls. The same IP address, 212.11.64[.]250, was observed running a "CyberStrikeAI" service banner on port 8080 and communicating with targeted FortiGate devices, with the campaign infrastructure last seen running the tool on January 30, 2026. The researchers warn that the tool's orchestration of over 100 security tools, usable by low-skilled operators, could accelerate automated targeting of edge devices such as firewalls and VPN appliances.