INTEL_REPORT
arXiv — Cryptography & Security (cs.CR) · published 5/21/2026, 4:00:00 AM · TLP amber
Summary
Ingested excerpt (first ~500 chars of normalized text).
Trusted Weights, Treacherous Optimizations? Optimization-Triggered Backdoor Attacks on LLMs arXiv:2605.20641v1 Announce Type: new Abstract: Inference optimization is a vital technique for deploying LLMs at scale. Compilation is the most widely adopted optimization technique for LLMs. While it assumes semantic equivalence between the original and compiled graphs, we first uncover its numerical side effects can be maliciously exploited to implant stealthy backdoors in LLMs. We…
https://arxiv.org/abs/2605.20641
sha256:cccfc220ebaf16b417159a55f93c5d06bbea8dff16867e3d1f76ff474321f448
What we pulled out
Deterministic extractor (IOC + allowlisted tokens + ATT&CK IDs present in DB).
Indicators
Linked with report → mentions → indicator. Values open the indicator workspace.
No indicators linked for this report.
Malware families
Allowlist token matches only.
Threat actors mentioned
Allowlist mentions — not a formal attribution verdict.
ATT&CK techniques
MITRE IDs referenced in text and present in local technique table.
CONTINUE INVESTIGATION
High-signal pivots without leaving the thread you started in search.
Browse the report corpus.