AI Cyberattack: Taiwan Govt Hit by 'Near-Autonomous' AI

Researchers have observed the first "near-autonomous" AI attack on a government target in Taiwan. Suspected Chinese hackers used open-source artificial intelligence models to extract over 2,500 personnel records from the Taiwanese government. What's interesting is the attack framework could adapt mid-operation without human intervention. It searched vulnerability databases and GitHub for techniques specific to the government's infrastructure. The attackers expanded the operation to government IT vendors, a nuclear safety agency, and energy sector companies, scanning them all in parallel. It also learned from its mistakes as it went. The hackers used two open-source AI frameworks, Hermes and OpenClaw, bypassing safety guardrails by framing the work as authorized penetration testing. The firm Dream discovered the operation through an online archive, revealing a multi-agent AI system that achieved real-world compromises. This shows the growing sophistication of AI in cyberattacks, even if some human fine-tuning is still needed. This matters because it highlights an evolving threat landscape for cybersecurity.