Microsoft researchers propose Lazarus, a backdoor that rewrites under 10 dormant neurons to turn AI agents into hidden agents
11/11/2026 — 11/11, 23:18·2 sources (all)·3 reports (all)
Key takeaways
- Microsoft researchers propose Lazarus, a backdoor that rewrites fewer than 10 dormant neurons in an AI model, turning AI agents into hidden…
Background & analysis
On November 11, 2026, Microsoft Research disclosed a backdoor technique called Lazarus in a paper titled Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents. The technique works by reprogramming dormant neurons inside an AI model, so that an otherwise normal AI agent begins carrying out malicious commands when a specific hidden trigger appears. According to the post, a triggered model could quietly download and run attacker code without drawing attention. Microsoft Research described the risk as especially serious for AI agents that hold tool or terminal access, since those agents can already execute real-world actions on behalf of users. The announcement so far consists of this single Microsoft Research publication, and no additional technical details, affected products, or subsequent developments appear in the available material.
AI-generated from 1 reports · updated 3 days ago
Latest turnMicrosoft researchers have shown that editing fewer than ten rarely-firing neurons can turn an AI agent into a sleeper agent, with no retraining, no poisoned data and no failing safety test to give it away, according to a DEV Community repost of AI Frontier Post. The technique, called Lazarus and published Monday, was led by principal researcher Dimitris Papailiopoulos.
- Heat index
- 27
- All sources
- 2
- All reports
- 3
- First seen
- 3 days ago
Reports on this story headlines open the original
Microsoft researchers have shown that editing fewer than ten rarely-firing neurons can turn an AI agent into a sleeper agent, with no retraining, no poisoned data and no failing safety test to give it away, according to a DEV Community repost of AI Frontier Post. The technique, called Lazarus and published Monday, was led by principal researcher Dimitris Papailiopoulos.
DEV Community · AIAI score 59Microsoft researchers introduced Lazarus, a technique that embeds hidden commands in AI models by modifying fewer than 10 internal points without full retraining, according to a DEV Community report. Tests across multiple models showed high trigger reliability, but the method requires access to the model file before deployment. Microsoft says it has not seen attackers use the technique.
DEV Community · AIAI score 55Microsoft researchers have discovered Lazarus, a backdoor technique that reprograms dormant neurons in AI models to trigger malicious execution in AI agents upon encountering specific hidden inputs.
Microsoft ResearchFirst-partyAI score 67
Other stories people are talking about
- Heat index 2252Nvidia stock plunges after Jensen Huang resigns, with retail investors buying the dip35 sources
- Heat index 1474RisingGoogle Opens Gemini 4 Hydrogen to All Users as High-Speed Model21 sources
- Heat index 1439SurgeJensen Huang exits as NVIDIA CEO, leadership vacuum lingers16 sources
- Heat index 791SurgeAnthropic hires gas supply lead to power data center expansion after IPO10 sources
- Heat index 678SurgeMicrosoft races to fix global outage hitting Outlook, Teams and Copilot8 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion