AI Stack Alert Medium TD-2026-0103 For CISO, AI Eng, AppSec, SOC / IR Unconfirmed AI-assisted
'Sleeper Agent' Tactic Hides Payloads in Open-Weight AI Models
Researchers describe a 'sleeper agent' technique that hides malicious payloads in open-weight AI models. According to the source, malware can lurk inside just 0.00001% of a neural network. The source does not detail detection methods or affected products.
Why it matters
Organizations that download and run open-weight models from public hubs may load hidden malicious code without noticing. Standard checks on model behavior may not reveal such payloads.
Do this
- Inventory open-weight models in use and record their source, version, and file hashes.
- Only load models from trusted sources and verify checksums against publisher-provided values.
- Run model files through model-scanning tools that inspect serialized formats before deployment.
- Isolate model-hosting environments so that a compromised model has limited network and credential access.
Signals
- NoCISA KEV
- 22/100Priority
- 40%Confidence
Evidence
“Malware can lurk inside just 0.00001% of a neural network. How, then, can any user tell it's there?”
Sources
Affected / related
Sleeper agent payload hiding in neural networks, Open-weight AI models
Published 2026-11-16 19:10 UTC · updated 2026-11-17 01:50 UTC · AI-assisted summary, checked against quoted sources · Unconfirmed: single non-official source · How we score