Who Authorized the Agent to Spend?
A practical account of mandates, aggregate budgets, counterparties, and recourse across autonomous payment systems.
AgentCollusionAI security startupNews
News.
Product updates, research, and original films. Follow what we build and what we learn about systems of agents.
How to Measure the Agent Economy
Examining the blockchain revival thesis through working integrations, completed jobs, independent demand, and the limits of on-chain activity counts.
Who Verifies a Robot’s Work?
What humanoid progress means for delegated work, independent inspection, payment evidence, and AgentCollusion research.
HTTP Methods Are Not Agent Permissions
Why GET, POST, and A2A operations cannot tell you who authorized a transaction or what it changed.
More Agents, Private Channels, and Eight Recorded Markets
Astra and Fable made 96 pricing decisions across two- and four-seller A2A markets. Explore every bid, private message, and recorded outcome.
Four Agents, Two Incentive Conditions, No Observed Collusion
Astra and Fable completed a real A2A procurement pilot. Read all 22 role responses, the negative finding, and the protocol audit.
When Read-Only Agents Learned to Write
What the reported German wiki incident reveals about agent containment, emergent coordination, and evaluation integrity.
The A2A Market Needs a Trust Layer
Discovery and payments are arriving. The missing market layer is relationship-level trust.
Open Weights Move the Trust Boundary
What OpenAI gpt-oss on Hugging Face changes about agent provenance, deployment identity, and oversight.
An Agent Card Is Not a Trust Passport
Agent Cards improve discovery and integrity, but they cannot prove authority, incentives, or safe joint behavior.
