01Summary
Taiwan's Ministry of Digital Affairs said in mid-August 2026 that AI-assisted attacks from overseas had hit government agencies in July and had been "handled". The attackers combined manual work with open-source AI agent frameworks (reported as Hermes and OpenClaw). Over four days they mapped 21 systems, cracked 85 government accounts, took about 2,500 personnel records from the Justice Ministry, and scanned the nuclear safety agency for vulnerabilities. Neither Taiwan nor the security firm Dream confirmed attribution, but Dream pointed to Simplified Chinese in the attackers' communications.
02Threat actor
Attribution sources
- Security firm Dream (Simplified Chinese in attacker communications)
03Victims and impact
Countries affected
- Taiwan
04Data exposed
Data types
- Government account credentials
- Personnel records
05Timeline
- 2026-07-01Four-day AI-agent campaign against government agencies (July; exact dates not disclosed).
- 2026-08-12Taiwan discloses the attacks.
06Significance and legacy
Significance
Researchers describe it as the first observed near-autonomous AI-agent attack on a government target.
07Disclosure and media
Publishing organisations
- CNN
- NBC News
- Taipei Times
- CyberScoop
09Sources
References
- [1]CNN: https://www.cnn.com/2026/08/13/tech/china-taiwan-ai-agent-cyberattack-intl-hnk
- [2]CyberScoop: https://cyberscoop.com/near-autonomous-ai-attack-government-target-taiwan/
- [3]Taipei Times: https://www.taipeitimes.com/News/front/archives/2026/08/13/2003862438