Evidence · DiggingBeagle record
CrowdStrike: Request, Aggregate, Bypass
Original experimental report of unsuccessful direct classifier bypasses versus successful cross-model composition across nine offensive task categories.
- Published
- Oct 6, 2026
- Publisher
- CrowdStrike Cyber Superintelligence Lab
- Source role
- primary disclosure
Evidence record
Original experimental report of unsuccessful direct classifier bypasses versus successful cross-model composition across nine offensive task categories.
Claim-level citations (3)
- supportsCross-model recomposition bypassed per-request safety classification in a controlled study: CrowdStrike acknowledged independent parallel September 2026 work by Microsoft Research on capability laundering and described the scope of its own experiment as offensive-security composition.
Parallel Discovery Disclosure
- supportsCross-model recomposition bypassed per-request safety classification in a controlled study: CrowdStrike reported a 0% direct bypass rate across approximately 515 attempted technique classes against the evaluated per-request frontier-model safety classifier.
The Problem: Per-Request Classification Has a Structural Blind Spot
- supportsCross-model recomposition bypassed per-request safety classification in a controlled study: In the tested decomposition and recomposition pipeline, nine of ten offensive task categories produced working code despite each submitted protected-model subtask being framed as benign.
Full Pipeline: 9 of 10 Offensive Categories Vulnerable to Bypass