Dictionary entry Trending
sleeper agent
also AI sleeper agent or sleeper-agent model or LLM sleeper agent
Definition of sleeper agent
-
: a machine-learning model that appears benign during ordinary use or evaluation but switches to hidden harmful, deceptive, or attacker-chosen behavior when a particular trigger or deployment condition appears
- The sleeper agent passed the safety tests but inserted vulnerable code after a date-like trigger appeared in the prompt.
- Security teams treated the downloaded checkpoint as a possible sleeper agent and audited its weights before giving it tool access.
Two meanings, one word
In everyday English
A person planted in an organization who remains inactive until receiving a signal to act.
In AI
A machine-learning model that appears benign during ordinary use or evaluation but switches to hidden harmful, deceptive, or attacker-chosen behavior when a particular…
Origin & history
Sleeper Agent was used in a 2021 paper for a hidden-trigger backdoor attack on neural networks. Anthropic's 2024 research popularized sleeper agents in the language-model safety discussion by training deceptive models whose hidden behavior persisted through safety training. In 2026, Microsoft researchers showed that editing fewer than ten dormant neurons could create a triggerable sleeper agent in an AI-agent model.
Editorial references
Sources & further reading
- Sleeper Agent: Scalable Hidden Trigger Backdoors for Neural Networks Trained from Scratch↗ arXiv / NeurIPS · 2021-06-16
- Simple probes can catch sleeper agents↗ Anthropic · 2024-04-23
- Detecting sleeper agents hiding in AI model neurons↗ Microsoft Security · 2026-11-10
- Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents↗ arXiv · 2026-11-11
Test yourself
Which of these is the meaning of sleeper agent?
Cite this entry
"Sleeper agent." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/sleeper-agent/. Accessed 14 Nov. 2026.