Know if your AI is still doing its job. Without opening the black box.
AITrustMeter continuously tracks the relationship between what your AI system is given and what it decides. No access to weights, code, or training data required. Just the inputs and outputs you already have.
Product demo
See the AI Trust Meter in action
A short walkthrough: upload a CSV, set your acceptable range, and click any point to see the records behind it.
Why teams use AITrustMeter
Everything you need to trust an AI system you didn't build
Most AI you rely on is licensed, not built in-house. You can't inspect it. You can still monitor it.
Zero-instrumentation scoring
Works from input/output pairs alone. No API access to model internals, no cooperation from the vendor, no source code required.
Runs anywhere, stores nothing
The free evaluator is lightweight JavaScript that runs entirely client-side, in your browser. No backend call, no account, no data stored anywhere — it works the same way on any platform, and nothing you type or upload ever leaves your machine.
Catch drift before it costs you
A statistical monitor built on Irene Aldridge's peer-reviewed research detects behavioral drift in as few as 11 periods after it starts, long before it shows up in a P&L or an audit.
Vendor accountability
Log which model version was active at every point in time, so a drift event can be traced to a specific upstream change instead of becoming a mystery.
Evidence, not just alerts
Every alarm comes with the underlying series that triggered it, so a human reviewer can actually investigate instead of just trusting a red light.
Your own thresholds
Calibrate sensitivity to your own risk tolerance. A trading desk and a compliance team don't need the same alarm settings, and now they don't have to share one.
Published, defensible methodology
Built on peer-reviewed covariance-based evaluation research, not a black-box scoring algorithm of our own that you'd have to trust blindly.
The method
Four steps, no black box required
Plug in AI Trust Meter
Connect the client-side AI Trust Meter app to the input/decision pairs your AI system already produces.
Get a live drift score
A continuously updated statistic shows how far current behavior has moved from your validated baseline.
Set your threshold
Decide how many standard deviations of drift should trigger a review or an escalation.
Review the evidence
Every alert ships with the exact series that triggered it, ready for a human to investigate.
Built by
Research-backed, not hype-backed
Irene Aldridge
FOUNDER, RISKAICENTER · AUTHOR OF SEVERAL BOOKS · ADVISOR TO REGULATORS & HEDGE FUNDS
AITrustMeter is built on Irene Aldridge's peer-reviewed, covariance-based methodology for evaluating AI investment strategies without access to model internals — distilled from three years of her latest AI research and over twenty years building mission-critical, low-latency systems for the finance industry. The same underlying statistic is documented in her SSRN research and in AI Governance for Institutional Readiness in Finance, co-authored with Steve Krawciw. View the full research record on SSRN.
Pricing
Start free. Upgrade for continuous real-time monitoring.
All plans include the core model drift monitor. Paid plans add persistence, alerting, and multi-system support.
Free
- One monitored system
- Manual drift check, on demand
- Community support
Pro
- One monitored system
- Continuous monitoring with email alerts
- Custom alarm thresholds
- Full alert evidence history
Team
- Everything in Pro
- 10 seats, shared dashboard
- Vendor model-version logging
- Priority support
Need SOC 2, custom integrations, or a multi-year institutional license? Talk to us about Enterprise →
Questions
Before you sign up
Do I need to give AITrustMeter access to my AI model?
No, and that's the point. You provide the inputs and outputs your system already produces. The model itself stays exactly where it is, under whoever already controls it.
What if I don't have a clean validated baseline to compare against?
AITrustMeter can help you establish one from a defined historical window, though a baseline you already trust will always give you a more meaningful drift signal.
Is this only for trading and finance?
The underlying method is domain-agnostic. Finance is the first vertical because that's where the research was developed and validated, but the same input/output covariance approach applies anywhere a black-box AI system makes repeated decisions.
Does this replace my existing model risk management process?
No. It's a continuous, automated layer that sits alongside whatever governance process you already have, specifically covering the gap between periodic audits.
Can I cancel anytime?
Yes. Pro and Team plans are billed annually with no long-term contract; cancel from your account settings and you won't be charged again at renewal.
Stop assuming. Start measuring.
The AI you're running today may not behave the way it did when you validated it. Find out, continuously, without ever needing to see inside it.
Get Started →