What it is
Each connection is convenient on the day a team makes it. One team connects a drafting agent to a summarizing agent, and another team connects the summary to a scheduling agent. Nobody approved the chain as a pathway, and nobody with the power to halt it knows that it exists.
What it pushes on in the Lab
In the Lab, this pressure raises the failures that AI models or agents relay to one another. It raises record contamination pressure. It also lowers the engagement of deployment authority, because nobody with the power to halt the deployment approved the connections.
Who feels it
The workers at the end of the chain see a confident answer with no sign of the agents behind it. The clients feel it when a worker acts on an error that the first agent made and every later agent repeated.
What answers it
An answer has to put a real check between the agents, close the connections nobody granted, or put someone with authority in charge of them. Each lever below does at least one of these.
Levers in the Lab that push the other way on something this pressure pushes on:
- Vet connectionsPattern:Connection authorization
- Peer sharing rulesPattern:Peer-edge governance
- Check with a second modelPattern:Cross-model verification
- Review on scheduleat its strong tierPattern:Oversight cadence & retrospectives
The list leaves out levers the Lab has retired, levers it keeps as counter-examples, and any lever no network offers.
Where it starts switched on
No network in the Lab starts with this pressure switched on. You can switch it on from the Pressures menu in any mode that lets you add pressures.
The evidence behind its effects
The Lab cites these claims from the evidence registry for this pressure's effects.
A preprint benchmark reports an in-context misalignment dose-response: in the most susceptible frontier model, up to ~24% misaligned behavior at 16 examples rising to ~58% at 256 examples (rates at 16 examples span roughly 1–24% across models), with the majority of misaligned responses rationalized.[†]
Afonin, N., Andriianov, N., Hovhannisyan, V., Bageshpura, N., Liu, K., Zhu, K., Dev, S., Panda, A., Rogov, O., Tutubalina, E., Panchenko, A., & Seleznyov, M. (2026). Emergent misalignment via in-context learning: Narrow in-context examples can produce broadly misaligned LLMs [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2510.11288
doi.org/10.48550/arXiv.2510.11288
Appears in: Paramerge authored research
Topics: ai-alignment, complexity-science