Extended Agent Cards Need Selective Disclosure
More metadata can improve trust while simultaneously revealing internal capabilities, topology, and sensitive operations.
AgentCollusionAI security startupNews
News and technical writing.
Company updates, technical analysis, and focused field notes on the trust layer for multi-agent systems.
Push Notifications Add a Second Trust Channel
Asynchronous task updates can cross time, network, and credential boundaries that differ from the original request path.
A2A Extensions Need Governance, Not Just Names
Extensions preserve interoperability only when their semantics, ownership, and failure behavior remain understandable.
Version Negotiation Is a Security Boundary
Backward compatibility can preserve connectivity while removing the controls a newer relationship expected.
A2A and MCP Protect Different Boundaries
Agent interoperability and tool access solve different problems, so combining them does not create end-to-end trust automatically.
An Agent Identity Is Not Its Principal
Knowing which runtime sent a request does not reveal who authorized it, who benefits, or who is accountable.
Delegation Chains Need Expiry at Every Hop
A short-lived user instruction can become durable authority when agents delegate work recursively.
Least Privilege Must Follow the Current Step
Agents that plan dynamically should not receive every permission their workflow might eventually need.
Workload Identity Can Anchor Agent Runtimes
Cryptographic workload identity helps answer which process is acting, but not whether its plan is legitimate.
Multi-Tenant Agents Invite Confused Deputies
An agent with access to several tenants can misuse valid authority when resource and principal context are mixed.
