A fund finds out a portfolio company is struggling when the founder finally calls, weeks after hiring froze and a co-founder quietly left. A company finds out its AI vendor changed model behind the scenes when a customer complains, not when it happened. Both are the same failure: risk that was visible in the underlying signals long before anyone was looking at it directly.
Sentinel Risk scores every private-market holding across five engines. Sentinel AI scores every deployed AI system across seven, then blends them into one composite. Neither product needs the other, but both land in the same place: a dashboard, a set of alerts, and a report that writes itself from real numbers.
Built for GPs, LPs, and family offices holding a book of private companies and funds. Every engine runs against the portfolio companies already on file, not a spreadsheet you have to keep uploading.
Adds up the whole book by sector, stage, geography, cheque size, and co-investor, and flags where the fund has more riding on one bucket than anyone decided on purpose.
Looks at every position's vintage, cheque size, ownership, and reported NAV against how much has actually been deployed, and surfaces which holdings are furthest from a realistic path to an exit.
Reads hiring patterns and key-person departures the way an experienced operator would, weighing how likely each signal is to mean a company is healthy, plateauing, or in real distress, calibrated against real 2022–2024 startup outcomes.
Runs the whole book through named scenarios, a 300 basis-point rate shock, a 2022-style credit contraction, an AI-driven SaaS multiple collapse, each with its own sector-by-sector haircuts, so a fund can see its downside before it happens instead of after.
Checks a GP-reported valuation against an independent illiquidity discount for that company's stage, from roughly a quarter off at seed down to nothing once a company is public, so a NAV that's drifted from reality gets caught instead of carried forward.
Built for CTOs, CROs, and compliance teams who need to know which of the AI systems they've deployed carries real exposure, not just which ones exist. Each system on file gets scored on all seven engines, then blended into one number leadership can actually act on.
Weighs the underlying model's real, measured hallucination rate against how sensitive the use case is, a credit-scoring or fraud-detection system carries far more exposure per hallucination than an internal chatbot.
Classifies each system the way the EU AI Act would, high-risk, limited-risk, or minimal-risk, by what it's actually used for, and estimates the real penalty exposure that classification carries.
Tracks how exposed a system is to its own provider, factoring in that provider's financial stability and its track record of deprecating models out from under the people building on them.
Rates prompt-injection exposure by what the system actually does, an autonomous agent acting on its own carries critical exposure; a low-sensitivity scoring model carries very little.
Projects real monthly spend from the model's own per-token cost and the system's actual daily user band, so a usage spike shows up as a cost risk before the invoice does.
Flags systems built on custom or fine-tuned models, or run on self-hosted infrastructure, where the real risk is how few people actually understand how the thing works.
Weighs how much of the decision is actually automated against how reversible a wrong decision is, a fully-automated, irreversible call carries far more exposure than one a human reviews first.
All seven engines roll into a single weighted score per AI system, so leadership gets one number to track per system, not seven to reconcile.
Every report pulls straight from the latest computed scores, concentration, exit risk, stress results, and valuation checks, so what's written matches what the dashboard already shows. Nothing is drafted from memory.
Risk conditions that cross a threshold raise an alert immediately, and every alert stays open on the dashboard until someone on the team actually dismisses it.
An admin can send the whole team a real digest email in one click: the current concentration and valuation-integrity scores plus the three most pressing open alerts, no dashboard login required to stay informed.
Every organisation on Sentinel is its own fully separate workspace. Nothing crosses between tenants, and access inside each one is scoped by role.
A private-market portfolio and an AI system stack fail for different reasons, but both fail quietly. Sentinel is built so neither one does.