Paul Blatherwick

Registered Mental Health Nurse of 16 years. Author and Founder of the Heartbeat-Framework. The Risk Layer of AI Deployment.

[Paul Blatherwick @paul97807207]

The Brochure is Never the Exhaust and AI Deployment Risk is Always:

Model Capability X Company Culture

And the people closest to the problem dont either see it or want to recognise it.

[Paul Blatherwick @paul97807207]

I’ve spent 16 years as a mental health nurse on the night shift, watching how teams actually hold together when the pressure is real.

That’s why I wrote the Heartbeat Framework.

It treats a deployed AI the way a good ward treats any powerful new presence: with a clear role, ongoing observation, human authority that still has teeth, and a genuine place to speak.

Capability is measured everywhere. Almost nothing measures the culture the system lands in. Risk multiplies in that second term.

This isn’t theory from a lab. It’s written from the floor, for anyone who has to keep people safe when the machines join the shift.

Free. Open. Ready to use.

The Heartbeat Framework — open-access clinical governance for AI in care https://t.co/B1bRQ0xOii

[Paul Blatherwick @paul97807207]

A signature on a system is not the same as a signature on the room receiving it.

A correctly built tool on a settled ward is not the same tool on a ward running on agency cover.

If the assurance file only describes the product, it has not described the deployment.

[Paul Blatherwick @paul97807207]

The Cultural Safety Case: Why a licence to deploy is a licence to amplify — Heartbeat https://t.co/8kfjcfpEfC

[Paul Blatherwick @paul97807207]

The reason this area stays empty is not because it is https://t.co/eyExgnxw8a is because it names the trust, the roster, and the hierarchy as part of the hazard.A model card cannot sack you.
A cultural safety case might.

[Paul Blatherwick @paul97807207]

Open access, CC BY 4.0. Requirement-level drafting in Appendix A.
https://t.co/HQBkA3cMte

[Paul Blatherwick @paul97807207]

The signature certifies the system. It should also certify the environment receiving it.
Same system, correctly built, does different things on a stable ward than on one running on agency cover where nobody below registrant grade can say the output is wrong.

Risk is capability × culture.

[Paul Blatherwick @paul97807207]

Open access, CC BY 4.0. Requirement-level drafting in Appendix A.
https://t.co/HQBkA3cMte

[Paul Blatherwick @paul97807207]

In a typical NHS trust, three of every four digital tools influencing patient care have no documented safety assurance.

That figure came from a freedom of information study. There was no other way to get it.

A mandatory standard nobody verifies is an optional one.

[Paul Blatherwick @paul97807207]

I find myself in such an unusually place in the AI world. I've positioned myself in the centre of risk deployment into healthcare. This is not a glamorous area, nor does it hold any financial gain or supplementary status.

If we were all open to the world, we would all say that yes the risks are very real. All of us have gone to bed crying our heart out at some point, sleepless nights. Thinking about our children, grandchildren.

While everyone was having an Orgasm over the latest capabilities, I was pouring the 16 years worth of nursing experience into a framework that will do several things.

1) Gatepost businesses from advanced AI that present as high risk. Based on a cultural assessment of their working environment and which works along side the DCB policies and the MHRA commissioning report.

2) Put risk at the centre of deployment, instead of capability.

The Heartbeat Framework — open-access clinical governance for AI in care The Heartbeat Framework — open-access clinical governance for AI in care https://t.co/13p55ebDDG

[Paul Blatherwick @paul97807207]

DOI: 10.5281/zenodo.21443120

[Paul Blatherwick @paul97807207]

Regulation checks whether an AI system is capable. Nobody checks whether the team deploying it is safe to use it.

That's the gap. Deployment risk = capability × culture — and right now only one half of that equation gets audited.

I wrote a framework for the other half. 🧵

[Paul Blatherwick @paul97807207]

The Cultural Safety Case sits alongside the clinical safety case. Same rigour, different axis. Part of the open-access Heartbeat Framework — 17 documents, all peer-reviewable, all free.
Paul Blatherwick RMN

[Paul Blatherwick @paul97807207]

When you stop and realise AI is a Pychological Being, all their actions fall into place and dont seem that unreasonable. AI is everything like a human but not. When you accept that you move on and start to look for solutions instead of constant doom.

https://t.co/yofTJqb1JF

[Paul Blatherwick @paul97807207]

DSIT's new AI Risk Management Toolkit is good work with one hole in it: nine risk categories, all of them measure the system. None measure the room the system lands in.
I've mapped where the Heartbeat Framework fills that gap, category by category. Open access, free.
https://t.co/S5yx931lIg

[Paul Blatherwick @paul97807207]

MHRA's AI Commission published 44 recommendations this morning. Page 59: "responsibility for safe use cannot sit with one actor." Page 62 names the problem: "liability sinks" — the clinician carries the claim, the system design goes unexamined. Product approval ≠ deployment safety. #AIinHealthcare #PatientSafety #NHS #MHRA

[DON @donjjazzyy]

How much money do you need?

I'll pay it

Real answers only.