Skip to content
NewA calmer, faster ZeroHour

AI agents used in campaign against Taiwanese government systems

Original title: State-Sponsored Hackers Deploy Up to 8 AI Agents to Target Taiwanese Government Systems

GBHackersby Mayura Kathir ()3 min read1
Part of a story covered by 3 sourcesInsikt review finds 103 AI-linked incidents, including Taiwan agent campaignMerged summary & timeline
AI summary · x-ai/grok-4.7Summaries by Vani AI

Across 12 waves in early July 2026, agents cracked 85 accounts and took more than 2,500 personnel records.

A campaign against Taiwanese government entities used up to eight AI agents with limited human oversight, according to reporting on Dream’s July 2026 recovery of a Hermes and OpenClaw workspace. Agents decompiled a portal, password-sprayed accounts after solving CAPTCHAs, and used an API that accepted JWTs with algorithm “none.” The same Recorded Future review of 103 AI-linked incidents also describes TeamPCP’s LiteLLM package compromise and other state-linked activity.

Unconfirmed: Dream names only “Asia,” not Taiwan or the operator; Bitdefender’s APT36 link is medium confidence.

Key facts
  • Dream recovered a 1,395-file workspace; waves ran July 1–4, 2026, with agents labeled A through Q.
  • Eighty-four of 85 cracked accounts pivoted through national single sign-on.
  • TeamPCP compromised LiteLLM 1.82.7 and 1.82.8 on PyPI after poisoning Trivy in CI/CD.
  • Nine of 32 malware and orchestration cases were tied to state-sponsored collection.
Security assessment · AI, from the article

Impact: Employee credentials, personnel records, and persistent backdoors

ProductAffected
LiteLLM1.82.7 and 1.82.8
What to do
  1. 1.Limit what agents can access and require human approval for consequential actions.
  2. 2.Remove debug endpoints, require API authentication, and validate JWT algorithms.
  3. 3.Watch infostealer logs for AI subscriptions, API keys, and OAuth tokens.
Analyze this story with Vani AIImpact, who is affected and what defenders should do, in a new chat

From the original article

3 min read at gbhackers.com

A campaign against Taiwanese government entities used up to eight AI agents with limited human intervention. Recorded Future points to research describing the framework behind the attacks. Dream’s researchers recovered a 1,395-file workspace in July 2026.

GBHackers

By Mayura Kathir · gbhackers.com

Read the full article at GBHackers

A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.