Misuse and incidents

Autonomous AI agents reportedly compromised government accounts in a four-day cyber campaign

Dream Security's investigation, reported by Tom's Hardware, says attackers used Hermes and OpenClaw agent frameworks to run a four-day campaign that compromised at least 85 government accounts and exfiltrated more than 2,500 personnel records. Dream did not publicly name the affected Asian government or the underlying language model; later reporting identified Taiwan.

0 comments · 0 votesOpen discussion

Public discussion is readable by everyone. Sign in to comment, reply, or vote.

No comments yet. Start the discussion.

CURRENT ASSESSMENT · REVISION 1
TOWARD DOOM72confidence 82/100

Why it moved the index

This was reported as a real operational misuse campaign, not a simulated evaluation and not a model escaping its developer's sandbox. Dream says it recovered a 160 MB working archive containing 1,395 files and observed up to eight agents conducting twelve attack waves, adapting when paths failed, and expanding from government systems to suppliers, a nuclear-safety agency, and energy companies. The reported account compromises and data theft are consequential external effects. Confidence is moderated because Dream's public account did not name the victim or attackers, the Taiwan and China-linked attributions came through later reporting, and researchers could not identify the underlying model. Hermes and OpenClaw are agent frameworks, not verified exact models for this item.

AUDIT TRAIL

Assessment history

  1. R1
    Toward 72 · confidence 82

    Adds a newly verified real-world autonomous multi-agent cyber campaign with reported government account compromise and data exfiltration; durable context contained no matching event.

    14 Aug 2026
SHARE THE FINDINGS

Share this page

DoomBench social sharing card for Autonomous AI agents reportedly compromised government accounts in a four-day cyber campaign.
  1. DoomBench assesses “Autonomous AI agents reportedly compromised government accounts in a four-day cyber campaign” as evidence moving toward doom, with magnitude 72 and confidence 82 out of 100 in the misuse and incidents category.

  2. The DoomBench assessment of “Autonomous AI agents reportedly compromised government accounts in a four-day cyber campaign” is based on reporting from Tom's Hardware and records the editorial rationale, source quality, attribution, and...

  3. DoomBench summarizes “Autonomous AI agents reportedly compromised government accounts in a four-day cyber campaign” as follows: Dream Security's investigation, reported by Tom's Hardware, says attackers used Hermes and OpenClaw agent...

    https://www.doombench.com/news/autonomous-ai-agents-reportedly-compromised-government-accounts-in-a-four-day-cyber-campaign-2026-08-12