#security
6 items
From Approval to Action: The Executor That Re-Binds to the Proposal
Most agent frameworks let the act step trust what the caller sends at execute-time, not what a human approved. That's prompt-injection-at-execute. The fix is one rule: the executor re-binds to the stored proposal — the caller passes only an ID.
Trust Through Boundaries: The Security Case for a Sovereign AI Workforce
Why a sovereign AI workforce you can govern is safer than an opaque cloud agent — boundaries, auditability, and isolation are the product.
Agentic Governance & Security
The emerging field of accountability, identity, and audit for AI agents — who authorizes an agent, what it did, and how you prove it.
Agentic Ops and Governance
Sensing demand for guardrails, tracing, and RBAC security in autonomous agent fleets — the operational standards for safe enterprise deployment.
Autonomous Agent Identity
The shift from shared API keys and service accounts to cryptographic, per-agent identity — what it means when machines outnumber humans 82 to 1.
Adversarial Gate Development: A Protocol for Building Substrate-Grade Software
We named the methodology we use to build SOS. It is not TDD, not security-at-the-end, and not code review. It is a formal protocol where correctness and adversarial review run as parallel gates before any feature merges into the substrate.