An Email the User Did Not Recognize
An anonymized case note on an unexpected email exchange, the boundary between drafting and sending, and the records needed to establish who authorized an action.
AgentCollusionAI security startupResearch
Research.
Experiments, technical notes, and public evidence on behavior that only becomes visible when agents are studied together.
How we work
Evidence before claims.
We study messages, incentives, tools, memory, authority, and joint outcomes as one connected system. Where possible, we publish the methods, limits, and evidence behind what we report.
Published work
Latest research outputs
NEC's AI Stress Checks Are Really a Test of the Workplace
NEC's AI employees report missing permissions and delayed decisions. What looks like an AI health room is an operational feedback loop.
Trump's AI Force: A Proposal Is Not an Operating Institution
Trump announced an AI Force and a future AI czar. The statement leaves practical questions about powers, funding and oversight unanswered.
Paper2Agent Gives Research Papers a Working Interface
New Nature coverage puts interactive paper agents in focus. Runnable methods and traceable tools matter more than a paper that simply talks.
Why Unit 42 Is Putting Multiple AI Models on Defense
Palo Alto Networks announced continuous frontier-AI defense on September 22. The key distinction is between finding an issue and verifying a fix.
When an AI Manager Makes the Report Worse
A paired experiment tests forced revisions in AI teams. Read the evidence, its limits, and watch our agent-perspective dramatization of the study shared by AIDB.