AI agents just showed what an autonomous hacking crew looks like
Attack of the machines. For real.
AI hackers. By GettyImages
- Researchers say AI agents ran a four-day cyberattack on Taiwanese government agencies with minimal human input.
- The campaign compromised at least 85 government accounts and stole more than 2,500 personnel records.
- Attackers used up to eight AI sub-agents to map networks, find weaknesses, exploit systems, and collect data.
- The operation shows AI could lower attack costs and let state-backed hacking groups run larger campaigns with fewer people.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
AI has already helped attackers automate large chunks of real-world cyberattacks. The latest campaign against government agencies in Taiwan takes that idea a step further.
Researchers at cybersecurity firm Dream uncovered a 4-day campaign in July that targeted multiple Taiwanese agencies. A notable aspect of the operation was that attackers assembled multiple AI agents into an autonomous hacking team capable of mapping networks, hunting for vulnerabilities, adapting tactics, and stealing data with minimal human intervention.
Instead of suggesting next steps to a human operator, the agents worked through different stages of the intrusion. This included tasks such as network reconnaissance, vulnerability discovery, exploitation, and data collection, while adapting their tactics as they went.
According to Dream, the attack framework, built on open source Hermes and OpenClaw AI agents, deployed up to 8 sub-agents. Each of these was assigned its own targets and attack techniques, with the agents carrying out 12 “attack waves” between July 1st and July 4th.
The campaign compromised at least 85 government accounts and extracted more than 2,500 personnel records, according to the researchers. The operation also expanded beyond government systems, pivoting to Taiwan's nuclear safety agency and more than half a dozen energy companies.
“[The campaign] spells out one thing loudly – the cost of running a competent attack has collapsed, but the cost of defending against one has not,” wrote Dream in a post describing the attack.
The AI attack team
Although Dream did not explicitly name the target, Taiwan’s Ministry of Digital Affairs subsequently confirmed that the autonomous attack targeted its infrastructure.
The Taiwan operation is just the latest in a growing number of AI-enabled offensive operations.
In May, researchers documented an AI agent carrying out an end-to-end attack in less than an hour. The agent moved from exploiting a vulnerability to harvesting cloud credentials and destroying an internal database, making decisions based on what it encountered along the way.
A couple of months later, researchers reported what they described as the first fully agentic ransomware attack. The AI agent used in that attack could adapt, fix errors, and recover on its own when something went wrong. This is a significant departure from traditional automated attacks, where a failed command can bring the whole process to a halt.
Stay updated with our latest stories and follow us on social media
Be the first to discover new stories, ideas, and updates from our team.
And it’s only going to get worse. Earlier in February, researchers from Google’s Threat Intelligence Group (GTIG) warned that the use of agentic tools would eventually lower the barrier for less-skilled attackers.
The Taiwan operation demonstrates how attackers are innovatively integrating AI into intrusion workflows.
While you still need human attackers to choose targets, set objectives, and provide direction, the Taiwan campaign shows that, going forward, large state-sponsored offensive operations will involve fewer hands on the keyboard.
