Skip to content
New kind of attack — caught and fixed before it was publicly known·Read the timeline
AI App Platforms

Your users generate thousands of agents. You can't manually review them — and you can't let one pop your cloud account.

Embed Probity as your runtime. Every generated agent boots in a sealed box, gated and signed automatically. You ship “secure by default” — with an independent attestation no host can issue about its own code.

The problem

Scale breaks manual review. One rogue agent is a headline.

Your platform deploys agents in minutes — you're hosting code you've never read. One that quietly exfiltrates a user's keys, phones home on certain inputs, or escalates under load: that's your incident report, your churn, your funding-round conversation.

Static scanners read source and guess. You need a verdict from execution — boot the real thing, throw known attacks at it, catch what only appears when it runs.

The self-attestation trap

You host the code. You can't grade your own homework.

Any trust signal you issue about agents you host is self-attestation — and a compliance officer won't accept it for SOC 2, ISO 42001, or EU AI Act review. Probity is the independent third party baked into your runtime: we sign the verdict, and your customer's auditor re-verifies it offline with a public key.

Why a host structurally can't attest to its own code →

One engine. Two moments.

The Gate is the X-ray: every generated agent gets scanned before it goes live. The Runtime is the quarantine: passing agents stay in the sealed box, every action gated and signed.

the Gate (pre-deploy)

Every generated agent gets a signed verdict before it touches user data.

Hook the Gate into your deploy pipeline: boot the agent in a sealed box, throw the known attack corpus at it, get a signed pass/fail in seconds. A fail never reaches production. A pass is a Trust Receipt — not a guess.

the Runtime (always-on)

Agents that pass live in the sealed box permanently.

Every production action runs gated, signed, and frozen the instant it steps out of bounds. Incidents go from “we think it was contained” to “here is a tamper-evident, signed record of exactly what it did.”

What your enterprise customer holds.

A Trust Receipt they can re-verify offline — independent of you, independent of us, readable with a public key and nothing else.

Offline re-verification — no Probity software, no network
$ python3 probity-verify.py ./verdict.intoto.jsonl probity.pub
signature OK · keyid 427a1092… · verdict PASS
tamper one byte → verification fails

Your customer's auditor runs this — the verifier and the pinned key are published at /verify, and the same check runs from a terminal or from CI. They hold the Trust Receipt; the verdict is deterministic, so it stands up in a compliance review or a procurement call.

Embed the runtime. Ship secure by default.

One integration. Every generated agent gated, signed, and independently attested — the trust signal your enterprise buyers need and you can never self-issue.

Get access

Run your agents through Probity.

A few details about your setup and we'll get you gating agents fast.