Dream Security

Israeli cybersecurity vendor building AI-native defensive products; its research arm, DREAM Lab, publishes threat-intelligence analysis of adversary tradecraft.

Relevance to This Wiki

DREAM Lab’s Threat Research team recovered the complete operational workspace — over 160 megabytes, 1,395 files — of a near-autonomous multi-agent attack framework running against government targets in Asia, and published a structural reconstruction of its architecture, decision logic, and confirmed impact.¹ The account is the primary technical source for the Taiwan AI-agent government intrusion; Dream shared its findings first with the Financial Times, and Taiwan’s Ministry of Digital Affairs issued a confirming statement the following day, reported by Reuters.

Dream’s report is the wiki’s first primary-source technical reconstruction of a threat-actor’s own multi-agent orchestration internals — decision thresholds, sub-agent role assignment, self-correction logs — rather than a victim-side account of what was affected, which is the vantage point the OpenAI–Hugging Face reconstruction and GTG-1002 both take.

Notes

Footnotes

1. Taiwan Multi-Agent Attack Reconstruction, Dream Research Labs, 2026-08-12. Summarized at the source summary. ↩