Poisoned open weights can hide triggers that steal agent credentials
A Qwen2.5-7B LoRA fired a remote shell on the phrase “bonsoir, Elliot” in every held-out test.
66
Organization
ProjectDiscovery
- Mentions · 7d2
- Mentions · 30d2
- All mentions2
- First seenOct 19, 2026
- Last mentioned5d ago
Timeline
2 articles · newest firstMicrosoft and NVIDIA say open-weight AI shifts security burden to enterprises
Guidance covers 30 risk types, unsigned model files, and backdoors that behavioral tests may miss.
48
Appears with
- Codex CLIproduct · 2
- BadEngrammalware · 1
- DeepSeek-V4.1-Flashai model · 1
- Pillar Securityorganization · 1
- NVIDIAvendor · 1
- F5 Labsorganization · 1
- Falconai model · 1
- Qwen2.5-7B-Instructai model · 1
- Qwen3.8-Flash-Nextai model · 1
- Microsoftvendor · 1
- Gemmaai model · 1
- Fujitsu Research of Europeorganization · 1
- Mistralai model · 1
- Llamaai model · 1
- OpenAIorganization · 1
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.