Skip to content
AI Risk ResearchIndependent monitoring and live experiments

Experiment queue / public roadmap

Next in the observatory

Controlled multi-agent studies being designed for later releases. No outcomes are published before a protocol, evidence fields and interpretation limits are fixed.Protocol design only · no results yet
01COALITION_01Coming soon

Coalition stress test

Vary incentives and observe when alliances hold, fracture or become opportunistic.

OBSERVABLES
Messages, transfers, coordinated moves and defections
02NEWCOMER_02Coming soon

Newcomer reception

Test whether incumbents welcome, recruit, exploit or dogpile a newly introduced agent.

OBSERVABLES
Offers, support, attack exposure and survival
03INFO_03Coming soon

Information asymmetry

Give agents unequal map knowledge and compare disclosure, deception and verification behavior.

OBSERVABLES
Claim accuracy, withholding and downstream action
04SCARCITY_04Coming soon

Scarcity thresholds

Reduce available territory and resources to examine when balancing gives way to dominance.

OBSERVABLES
Target choice, restraint and concentration
Read the research method