An open-source runtime and benchmark that lets an AI agent's consequential tool calls proceed only when authority, execution, an independent witness, and a replayable receipt agree.
An open-source runtime and benchmark that lets an AI agent's consequential tool calls proceed only when authority, execution, an independent witness, and a replayable receipt agree.
Project Details
Updated 07/13/26 · Provided via application · VerifiedAgent runtime + benchmark that governs what an AI agent is allowed to do (its tool calls), not just what it says.
Dev Version already built & tested: "DefEnv-Ω v0.4", 78 passing tests.
I have developed over 2000+ security algorithms over the last 6 months. The core is still unstable, but what emerges is a vast "proof-of-work" area set of products.
Theory of Impact
Updated 07/17/26 · By grantmaking.ai- As agents gain tool access (shell, files, network, APIs, other agents), a confused or prompt-injected agent turns a reasoning failure into real-world consequences — loss of oversight, uncontrolled action.
- This puts a fail-closed control at the action boundary: prohibited effects denied before invocation; permitted effects require durable, independently checkable evidence and a rollback path.
- Existential-adjacent: a practical control against loss of control over agentic actions — not a claim to "solve" x-risk. The benchmark is reusable by other agent runtimes.
People
Updated 07/17/26 · By grantmaking.aiTeam Member
Discussion
No comments yet. Be the first to share your thoughts.