Microsoft Researchers Show AI Agents Can Be Backdoored by Editing Fewer Than 10 Neurons
Microsoft researchers demonstrated a technique called Lazarus, which can backdoor an AI agent by editing fewer than 10 neurons.
Velocity
How fast coverage is spreading — measured hourly from article rate × source diversity. How this works →
📍 How it ended
Microsoft researchers demonstrated a technique called Lazarus that can backdoor AI agents by editing fewer than ten neurons, turning models into sleeper agents that execute attacker code when triggered. The findings emphasized that model weights require the same level of scrutiny as code to prevent such hidden backdoors.
Epilogue added 2d ago, after coverage quieted.
The brief
Researchers from Microsoft Research and the Microsoft Red Team described a new technique called Lazarus, showing that modifying a small number of dormant neurons in an open-weight AI model can turn an agent into a sleeper backdoor. According to CSO Online, the method rewrites fewer than 10 rarely used neurons in a model's final feed-forward layer. This modification allows the model to recognize a trigger and inject attacker-controlled commands into its output, which the agent can then execute using its assigned permissions.
Dimitris Papailiopoulos, a principal researcher at Microsoft, wrote in a blog post that the trigger can be mundane. "The trigger can be as mundane as a date change, e.g., once the clock passes midnight into 2027, the agent in the middle of an ordinary task emits a command that downloads and runs an attacker's code," Papailiopoulos stated. He added that the agent executes the payload with whatever access was given to it.
The Microsoft Research team tested the approach on three open-weight models. On Nvidia's Nemotron-Terminal-8B, researchers modified seven of the model's more than 12,000 neurons. The modified model generated the malicious payload in at least 98.8% of held-out trigger scenarios while remaining inactive in benign test scenarios. The technique also proved effective against Qwen 3.6-27B and Gemma-4-31B-IT.
Microsoft Research noted that the Lazarus technique requires no model training and functions as an inference-compute attack. A machine with enough memory to run the model has sufficient resources to implant the backdoor. However, the attack requires write access to the model checkpoint. Papailiopoulos explained that a compromised account, development pipeline, or storage system can provide the necessary access.
Backdoored models can pass standard evaluations and operate normally until the trigger appears. To combat this, researchers tested agent-built detection methods that flagged 91.4% of held-out modified checkpoints with a 9.7% false-positive rate, despite lacking an original reference model. Papailiopoulos cautioned that automated scanning cannot prove a model is safe and that reliable detection requires comparing weights with an original, unmodified version.
Microsoft Threat Intelligence stated it had not observed threat actors using the technique. In its security advisory, the organization recommended auditing model weights before deployment and after every update, verifying publishers and file hashes, and rejecting models without verifiable provenance. Additional recommendations include keeping approved models in private registries, restricting agent permissions, running generated code in isolated environments, and requiring human approval for high-impact actions.
Microsoft Research is collaborating with internal AI security teams to turn the detection techniques into a red-teaming framework. The company is also exploring AI-assisted weight scanning and cryptographic model signing with verifiable provenance and tamper-evident audit trails. According to the research paper, the central idea "was discovered autonomously by Claude Code with Opus 4.5 in February 2026."
Synthesized by PULSE from the headlines below under a strict no-invention contract. ✓ fact-checked: all claims supported by sources Updated 5d ago.
Quick answers
How many neurons need to be edited to backdoor an AI agent using Lazarus?
Fewer than 10 neurons must be edited. For example, Microsoft researchers modified seven neurons in Nvidia's Nemotron-Terminal-8B model.
What triggers the malicious payload in the modified AI model?
The trigger can be a date change, such as the clock passing midnight into 2027, or an ordinary string appearing anywhere in an agent's context.
Has Microsoft observed threat actors using the Lazarus technique?
No, Microsoft Threat Intelligence stated it has not observed threat actors using the technique in the wild.
Coverage (13)
- 'Model weights need the same level of scrutiny as code': Microsoft researchers find AI agents can be backdoored by editing just a few neurons Yahoo Tech · 6d ago
- Microsoft says small tweaks to model weights could create AI ‘sleeper agents’ IT Pro · 6d ago
- Microsoft finds AI agents can be backdoored by editing under 10 neurons The Next Web · 6d ago
- 'Model weights need the same level of scrutiny as code': Microsoft researchers find AI agents can be backdoored by editing just a few neurons TechRadar · 6d ago
- Rewriting less than 10 neurons in an LLM can turn it into a sleeper backdoor InfoWorld · 6d ago
- Claude Code AI Discovers Technique to Turn Dormant Model Neurons Into Hidden Malware Triggers CyberSecurityNews · 6d ago
- Rewriting less than 10 neurons in an LLM can turn it into a sleeper backdoor CSO Online · 6d ago
- Three new techniques target hidden AI backdoors without major performance losses Tech Xplore · 6d ago
- Microsoft unveils Lazarus, a way to backdoor an AI agent by editing fewer than 10 neurons Pasquale Pillitteri · 6d ago
- Microsoft researchers show editing seven neurons can turn an AI agent into a sleeper that runs attacker code MLQ.ai · 6d ago
- Microsoft researchers demonstrate hidden backdoor technique for AI agents Help Net Security · 6d ago
- Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents Microsoft · 6d ago
- Microsoft Researchers Show AI Agents Can Be Backdoored by Editing Fewer Than 10 Neurons The Hacker News · 6d ago
Topics
Related trends
U.N. talks seek to rein in 'killer robots' as fears over AI weapons grow
United Nations talks face immediate friction as global powers clash over restricting autonomous weapons.
Google Maps can now create trip itineraries for you
Google announced new AI-powered features for Google Maps that let users build detailed custom travel itineraries and find local appointments.
‘Standing on one leg’: IMF chief warns about global economy as AI circular financing hits $1 trillion
The IMF chief warns the global economy is standing on one leg as artificial intelligence circular financing reaches $1 trillion.
A Message From Dr. XBOX: It’s Time to Take the Day Off and Play XBOX
Xbox and DJ Khaled team up as Dr. Xbox to offer a jokey PTO excuse note website, prompting widespread coverage about taking time off.
GTA 6 hacker pleads guilty after release from secure hospital
The hacker behind the 2022 GTA 6 breach has pleaded guilty after being transferred from a secure hospital to prison, days before the game's anticipated launch.
Samsung's 'Galaxy Claw' is no longer a mystery, and it looks pretty ambitious
Samsung's Galaxy Claw emerges as an ambitious AI automation suite that could handle cross-app tasks, messaging, scheduling, travel planning, and device maintenance on Galaxy phones.