Independent security and privacy reviews.

We review the same kinds of systems we write about. Each links to our research on it.

• Private-inference and attestation setups We review whether your users’ software checks the server’s attestation (its signed hardware evidence) by default, and what it does when that evidence is stale or altered. See Private AI verification.
• AI tool access and MCP servers We check which hosts and secrets an agent’s tool code reaches, and whether that matches its documentation. See MCP source study.
• Agent and wallet signing and approval paths We check who can sign and what each approval actually names. We also check whether the action a person approves is the one that runs. See Developer tool attacks and Agent trust.
• Consumer health data outside HIPAA We review where a health record leaves HIPAA, the US health-privacy law, in your product, and which rules and promises cover it after that. We read policies and data flows. A review isn’t legal advice. See Health records.
• Prompt-injection exposure in agent workflows We check which documents, emails or logs your agent reads could plant an instruction that adds a step, destination or permission, and what stops it. See Prompt injection and AI defense.

You can stop after any stage. Whatever we’ve found by then is yours.

• Scope check, 1 week We read your docs, map the code and make sure our conflict policy lets us take the work. Then you get a one-page plan covering what we’d check, what we’d need from you and what we’d leave out.
• Focused review, 3 weeks One system and one question from the list above. For example, does your users’ software check attestation before it sends anything? You get a written review.
• Full review, 6 weeks Up to three questions from the list, across one product, in one written review.
• Re-check, 2 weeks After you fix things, we test the same findings again and tell you which still hold.

The clock starts once we have the access the plan asks for. After the scope check, we give you a fixed price in writing.

You get a written review of what we looked at and what we didn’t. Every finding shows where we saw it, how serious we think it is and what would change our mind. The review ends with the limits of the check.

It isn’t a certification, and it won’t claim to verify your system.

We don’t run penetration tests, give legal opinions or issue compliance reports and certificates, such as SOC 2 or ISO 27001. If you need one of those, we’ll say so at the start.

We take no paid work from a company while our published research scores its product. Read our full conflict policy.

Your review stays private. We never publish it, its findings or your name without your written consent.

The people who’d review your system work here. We don’t publish their names, but you can meet them under NDA before we start.