Researchers observe first ‘near-autonomous’ AI attack on government target in Taiwan

Summary

Researchers have identified a cyberattack against the Taiwanese government that utilized open-source AI models, marking a significant step towards autonomous offensive operations. The attackers' framework adapted mid-operation, autonomously researching vulnerabilities and expanding its scope to include IT supply chain vendors and energy companies, while also correcting its own mistakes. This incident highlights the growing trend of threat actors leveraging AI for sophisticated cyberattacks.

Full text

Suspected Chinese hackers used open-source artificial intelligence models to run a cyberattack against the Taiwanese government in the first publicly known case of an autonomous AI hack hitting a government target, according to research published Wednesday. The hackers extracted more than 2,500 personnel records, among other data, in the “near-autonomous attack,” researchers at Israeli cyber firm Dream wrote in a blog post. The attackers set up the framework so that it could “adapt mid-operation without human intervention.” The framework “implements dedicated research phases it calls ‘Learning Cycles’ — autonomous sessions where the AI system searches vulnerability databases, GitHub repositories, and security research publications for techniques specifically applicable to its target government’s infrastructure,” the post reads. And then it kept going. “The attacker didn’t stop at primary targets,” Dream said. “It expanded the operation to government IT supply chain vendors, a nuclear safety agency, a government email system, and 7+ energy sector companies — scanning them all in parallel for misconfigurations, exposed admin interfaces, and exploitable vulnerabilities.” It also learned from its mistakes as it went on, Dream said in identifying what stood out about the campaign. Autonomous AI-powered cyberattacks have raised alarms in the past: Anthropic reported last fall that it stopped the first autonomous cyber espionage campaign, although researchers noted that the “autonomous” campaign still required significant human work. The Financial Times first reported the Dream research and details on the target. The hackers used two popular open-source AI frameworks, Hermes and OpenClaw, to set up the Taiwan operation. They bypassed safety guardrails by framing the work as authorized penetration testing, according to Dream. The firm discovered the operation via an online archive of 160 megabytes and nearly 1,400 files, revealing “a multi-agent AI system that achieved confirmed, real-world compromises against state infrastructure.” As with the autonomous cyber espionage campaign uncovered last fall, the attack Dream examined also noted the need for human tinkering. “We increasingly see threat actors leveraging AI for autonomous offensive operations,” the company wrote. “But building a system that actually works at this level takes more work than ‘just’ running a model. It demands careful adjustment to the specific task, optimization of agent coordination, and fine-tuning of decision logic — the kind of sophistication evident in this framework’s Bayesian prioritization, self-correction loops, and adaptive research cycles.” Share Facebook LinkedIn Twitter Copy Link

Entities