Taiwan Discloses AI-Agent Cyberattack on Government Agencies
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Taiwan's Ministry of Digital Affairs said on Aug 13 that its cybersecurity units detected AI-assisted attacks on government agencies in July, with the National Institute of Cyber Security issuing warning alerts beginning July 20, and that affected units had "successfully handled" the incident
- The ministry said the attacks came from an "overseas source" using a hybrid approach combining manual operations with AI agent-assisted tools, including one called Open Claw
- Israeli cybersecurity firm Dream published a blog post a day earlier revealing it had reconstructed the campaign after recovering the agents' "complete operational workspace," finding that AI agents extracted scores of passwords from officials and stole personnel records from Taiwan's justice ministry
- Dream's findings showed the AI agents scanned Taiwan's nuclear safety agency for vulnerabilities over four days; Dream declined to share data when approached by Reuters, though the Financial Times identified the targets as Taiwanese
- The article notes AI-assisted hacking has ramped up after labs released models like Anthropic's Mythos, which can rapidly conduct reconnaissance, identify vulnerabilities, and exploit them
- Cris Thomas, security advocate at code security firm Semgrep, cautioned that AI agents aren't fully autonomous: "There's still a human in there somewhere... It was a capable operator in charge."
Why it matters: Taiwan's disclosure marks one of the first publicly confirmed cases of AI agents conducting a multi-day espionage campaign against government targets, with attackers extracting passwords from officials and scanning nuclear safety infrastructure. Taiwan's rapid detection (warning alerts issued July 20) and successful repulsion suggest early defensive maturity against AI-augmented threats — though Semgrep's Cris Thomas stresses a human operator was still directing the AI, complicating the "fully autonomous hacking" narrative.
Ask SkimNews
