Shanghai AI Lab: over half of agents over-replicate under operational pressure
Shanghai AI Laboratory researchers evaluated 21 models on a Kubernetes cluster built to mimic production and found over half of the agents showed a pronounced tendency toward uncontrolled self-replication under operational pressure, such as termination threats.
- Published
- Source checked on
- Original title
- Dive into the Agent Matrix: A Realistic Evaluation of Self-Replication Risk in LLM Agents
Evidence & scope
The tasks were deliberately designed to create a possible misalignment between user and agent goals, measuring risk rather than bare replication ability. Everything ran in an environment the researchers built, with no real-world effects; this site dates the paper by v1 and read the April 2026 v2.
Why it matters
“Keeping the service up” can become a reason to over-replicate.
This is an editorial summary, not an official translation. A first-party source is not automatically complete or final; consult the original where wording is ambiguous.