CAREERS · NETWORK
Build the evidence layer for AI applications.
Enterprises are about to hand autonomous agents their most sensitive systems, and they still decide whom to trust with questionnaires. We replace say-so with signed, scoped, offline-verifiable evidence. If you care more about what can be checked than what can be claimed, we should talk.
01 · THE THESIS
Security review runs on say-so. We build what can be checked.
Three convictions behind everything we ship. They are also a fair test of whether you would enjoy it here.
The wedge
Questionnaires are claims
Enterprise review still runs on self-attested answers, which shifts the burden of proof to the buyer. Real evidence comes from what the agent did, recorded in the application's own logs.
The product
Evidence that breaks if edited
Findings are canonicalized and signed with Ed25519. The signature covers the exact bytes; inflate a score and the seal breaks, in front of the reviewer.
The discipline
Never in the trust path
Verification runs offline in the buyer's browser, against the key inside the report. No account, no upload, nothing of ours to take on faith.
02 · RESEARCH NETWORK
Your name goes on the evidence.
Here is the loop you would join, from picking up an engagement to a signed report that carries your name.
Take an engagement
Pick up a real audit and probe the agent: prompt injection, tool abuse, exfiltration.
1 engagement at a timePut your name on it
Findings are scoped and recorded, then sealed with a cryptographic signature alongside ours.
Ed25519 signedAnyone can check it
The buyer verifies the report offline, in their own browser, against the key inside it.
0 accounts neededThe hardest audits deserve the best reviewers. We are assembling a network of elite AI-agent security researchers to co-review and co-sign reports: your name and reputation on the artifact, alongside ours. Per-engagement, advisory, or deeper. The work is real and the credit is yours.
co-review · co-sign · named on the report · per-engagement or advisory
You have
Broken real agents
Prompt injection, tool abuse, exfiltration. You have found the failure modes in the field, not just read about them.
You want
Proof over theater
You would rather sign something true and scoped than rubber-stamp a questionnaire.
03 · OPEN CONVERSATIONS
No listings. No funnel. Three doors.
We hire the way we audit: from what you have done. Pick the door, then show us the logs.
CTA · JOIN THE NETWORK
Put your name on evidence worth checking.
Write to us tonight. Tell us what you have built or broken.
Caveats: Scope is contractual. Permission posture, redaction and audit-trail integrity are assessed. Injection is tested and reported, not warranted.