Rewriting fewer than 10 neurons can backdoor an open-weight LLM
Microsoft’s Lazarus edit of seven neurons in Nemotron-Terminal-8B made agents emit commands on an ordinary trigger, and no real-world use is reported.
58
AI model
Gemma-4-31B-IT
- Mentions · 7d2
- Mentions · 30d2
- All mentions3
- First seenSep 16, 2026
- Last mentioned8h ago
Timeline
3 articles · newest firstMicrosoft shows AI agents backdoored by editing fewer than 10 neurons
A trigger can make the model emit a command an agent may run; Microsoft has not seen attackers use the method.
56
Knowledgator Releases GLiFormer: A 575M-Parameter Encoder That Hits 91.10 F1 on Nested JSON Extraction Without Generating Tokens
Knowledgator released GLiFormer, an Apache-2.0 encoder (264M/575M) handling NER, classification, relations, and nested JSON extraction, scoring 91.10 F1.
35
Appears with
- Nemotron-Terminal-8Bai model · 2
- Microsoft Red Teamorganization · 1
- Knowledgatorvendor · 1
- ProjectDiscoveryorganization · 1
- Microsoft Foundryproduct · 1
- Microsoft Threat Intelligenceorganization · 1
- GPT-5-miniai model · 1
- GPT-5.6-lunaai model · 1
- Microsoft Defender for Cloudproduct · 1
- Microsoft Researchorganization · 1
- GLiFormer Large v1ai model · 1
- GLiFormer Base v1product · 1
- GLiFormer Base v1ai model · 1
- Microsoftvendor · 1
- Knowledgatororganization · 1
- Qwen 3.6-27Bai model · 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.