Skip to main content

Calibrated honesty

The AI Reliability Traffic Light

In a market full of hype, here's the boring, honest version: what AI is actually ready for today — and what it isn't. Updated regularly, with our reasoning and the date we last checked.

The legend

AI Reliability Traffic Light

AI reliability: Go
Reliable for this task when used as directed.
AI reliability: Amber
Workable — verify the output and keep a human in the loop.
AI reliability: Stop
Not reliable for this task yet — avoid it for now.

How we decide: evidence, not vibes

Each entry carries a candid rationale, the evidence behind it, and the date we last reviewed it. We review the register monthly, and we change a rating the moment the evidence does. When we're unsure, we say so. Nobody else will give it to you this straight — which is exactly why we do.

The register

The register is being prepared — check back shortly, or get the ratings weekly in The Signal.

How we keep this current

The register is a living document, owned by our AI Reliability Lead. On the first working day of every month they re-read each entry, re-check its sources against any material model or regulatory change, and re-stamp the review date — even when the rating doesn't move — so the "as of" date you see is always honest. A rating changes the moment the evidence does, and the bigger shifts get explained in The Signal.

Get the changes the moment they happen.

We update this register as the evidence moves — and explain what matters in The Signal, free, every week.

    We use cookies.