AI-assisted assaults accessed methods or retrieved information at 10 organizations, in response to a November 12 investigation by Palo Alto Networks’ Unit 42. The safety analysis staff tracked 25 operations towards at the least 19 organizations throughout August and September 2026, together with semiconductor firms, protection and coverage organizations, and civil-society teams.
Unit 42 mentioned the attackers used a custom-made model of Cairn, an open-source framework for AI-driven penetration testing, paired with Zhipu GLM and DeepSeek fashions.
The reported outcomes went past reconnaissance. The researchers reported unauthorized entry to a semiconductor firm’s manufacturing system and to a couple of million member information at a US lecturers’ union.
Attackers additionally modified a dwell standards-ballot doc and obtained nonpublic information at an business requirements group, the researchers mentioned.
People set targets, brokers sustained the work
People provided targets and objectives, typically including intelligence gathered beforehand. The framework then deliberate, executed and recorded subsequent work.
Particular person operations lasted from 20 minutes to greater than 50 hours, Unit 42 reported, with human steering nonetheless doable.
People selected targets and will hold steering whereas brokers repeated the intrusion work.
Marketing campaign outcomes are Unit 42’s findings. The diagram reveals the design, not measured human-intervention frequency. Operations and organizations are totally different items; no success fee is inferred.
That association let people direct the marketing campaign whereas brokers sustained the intrusion work. Unit 42 assessed {that a} single operator or small staff probably ran the exercise and argued that autonomous orchestration expanded the variety of simultaneous makes an attempt they might maintain.
Unit 42 concluded that AI elevated the size and attain of the marketing campaign, not its technical sophistication.
The November 12 investigation provides quantified outcomes to earlier reporting on AI-assisted assaults. CryptoSlate’s November 5 protection of CERT-In’s CRAFTY FALCON advisory already named Cairn amongst AI-assisted offensive-security instruments. The experiences describe totally different noticed exercise; shared tooling doesn’t set up a hyperlink between their operators.
Unit 42 assessed with reasonable confidence that the exercise it tracked represented China-nexus intelligence assortment, primarily based on language, tooling and goal choice.
For defenders, the report’s sensible warning is that acquainted gaps can help extended, coordinated intrusion makes an attempt. Unit 42 says identification hygiene, decreasing uncovered assault surfaces and software program patching stay efficient.
