Preliminary research
NemoClaw drive-by agent hijacking: bind-dependent defenses and persistent template poisoning
Drive-By Agent Hijacking: One Website Visit, Persistent Model Poisoning
AI-collected research leads through 22 September 2026, including targeted additions between broader sweeps. Unranked, incomplete, not community-vetted, and subject to change.
Connects a container integration’s non-loopback Ollama binding to disabled host checks and DNS rebinding. Modifying a model’s rendering template persists injected instructions where an earlier system-message change failed. The case demonstrates template precedence and deployment-dependent defenses without proving unrestricted downstream compromise.
Record
- Document
- Drive-By Agent Hijacking: One Website Visit, Persistent Model Poisoning
- Researcher
- Elad Luz and Ofek Itach
- Published by
- Cyera Research
- Date
- Topic
- Injection
In the archive
Tags
This page is the archive's own catalogue record. The research is the work of Elad Luz and Ofek Itach, first published at the original source. Preserved copies are kept so the citation survives its host; this one was last captured on .