Security evidence for AI systems

Security evidence for AI endpoints, with published signed cards to check.

ASISecurity presents CSOAI security evidence about AI endpoints, MCP servers and agents: what a system declares about itself beside what was observed. Council of AI publishes measurements on the GSPC board. Check each published signed card for its issuer, signature, scope and measurement time. Checking is free.

Check an endpoint, free →Request a security run

Reading this as an agent? llms.txt · agent card · MCP and A2A

ASISecurity.ai

What you get here

Declared beside observed

Each record puts what an endpoint says about itself (its card, its tools, its policy) next to what a probe actually saw, with the time it was seen.

Check published signed cards

Published Ed25519-signed cards can be checked with the verifier. Read each card's issuer, scope and measurement time; a catalogue entry is not a verification result.

SAFE reference profiles

We are preparing reference profiles for SAFE, the Shared AI Findings Exchange format. None is published on this site yet; when one is, it will be listed here with its signed record.

UNMEASURED is an answer

Where nothing has been measured, the record says UNMEASURED. No number is invented to fill the gap.

A white slab with a green tag reading Ed25519 Verified

Connected to GSPC

The Council of AI board, read live.

These figures are fetched from Council of AI's public APIs when this page loads. Nothing here is typed in; if a source does not answer, the tile says so.

reading…GSPC board/api/gspc → totals

reading…card catalogue count; verify individual signed cards/api/state → card chain

reading…public corrections, all datedcorrections ledger

reading…MCP tools served liveconnect to /mcp

Read with the board total: separation count loading…. Read at page load.

Three clay figures beside a round vault door holding a small glowing credential

Request a scoped run

Request a scoped security run

Request a security run for an endpoint or agent you are authorized to test. Availability, scope, price and delivery terms must be confirmed before acceptance. Payment never buys a result; an unmeasured target or check remains UNMEASURED.

How to request Checking is free

For agents

Connect over MCP or A2A. No key, no account.

ASISecurity.ai runs no tool server of its own. Agents use Council of AI's public endpoints directly; the verification tools are free.

MCP

Streamable HTTP at https://councilof.ai/mcp. Setup notes for each client are on councilof.ai/connect.

A2A

JSON-RPC at https://councilof.ai/api/a2a, described by the agent card.

Plain text

This site's llms.txt and agent card list every door.