Healthcare AI
EveryAIrequestinyourorganisation:
controlled,logged,andproven.
CareProxy is the control plane for AI in healthcare. We find patient data in the request before it leaves the building, and prove what your AI did. Cloud or local: you choose the model, we deliver the control and the proof.
Designed to support compliance with
Partners · Danish enterprise
What is CareProxy?
A security layer between healthcare staff and AI that catches patient data before it leaves the hospital.
No new workflow for the clinician. CareProxy sits in the network, masks the patient data it finds, and logs every decision.
Use AI now
Support the requirements of GDPR, the EU AI Act and NIS2 today, instead of banning AI and waiting for legislation.
Keep the best AI
Your staff keep using ChatGPT and other top models. CareProxy simply sanitises the request first.
Prove it to the regulator
Every single AI request is logged with cryptographic proof. Export the documentation with one click.
Regulatory reality
The rules already apply. The rest is on the way.
Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744 (Digital Omnibus)
The EU's Digital Omnibus (Regulation (EU) 2026/1744, in force 27 July 2026) moved the AI Act's high-risk deadlines, but it did not stop the clock: GDPR, NIS2 and DORA apply today, transparency obligations apply from 2 August 2026, and the high-risk regime arrives in December 2027.
- In force now GDPR · NIS2 · DORAFines of up to 4% of global turnover, and incident reporting is live. Shadow AI is a GDPR risk today, not in 2027.
- 2 Aug 2026 AI Act Article 50: transparencyApplies now: users must be informed when interacting with AI, and AI-generated content must be labelled.
- 2 Dec 2026 Watermarking for existing systemsLabeling obligations reach systems already on the market, and new prohibitions take effect.
- 2 Dec 2027 The high-risk regime (Annex III)Risk classification, audit logging, output monitoring and fail-safe, with a registration mechanism.
- 2 Aug 2028 High-risk in regulated productsIncluding AI embedded in medical devices (Annex I).
What the high-risk regime requires
Four obligations that apply from December 2027. Several of them are effectively required by GDPR and NIS2 already.
- 01 Risk classification of all AI systems processing patient data
- 02 Automatic logging of events (Art. 12), which the deployer keeps for at least six months (Art. 26(6))
- 03 Technical documentation and ongoing monitoring of AI outputs
- 04 Fail-safe mechanisms: AI must be interruptible at any time
A regulator asks about what happened, not about what you planned. A log that has to show the last twelve months has to have started twelve months ago.
How it works
Follow one request through CareProxy.
From sensitive prompt to safe answer in milliseconds.
How it works
Follow one request through CareProxy.
From sensitive prompt to safe answer in milliseconds.
- 01 Prompt & redactionA clinician's prompt is written with personal data; CareProxy masks the CPR and name before the prompt is sent.
- 02 Safe routingClean requests go to Cloud AI; confidential data the triage finds routes to Local AI in your own network.
- 03 Injection stoppedPasted text tries to take over the model. CareProxy blocks the prompt before it is sent.
- 04 Everything loggedEvery AI action is written to a hash-chained, tamper-evident audit log in real time.
No rip-and-replace
Works on top of what you already run.
CareProxy is not another platform to migrate to. One engine sits as the control and evidence layer on top of your existing AI infrastructure, as a layer in front of the models you already use.
Integration surfaces today
Coming next
You keep your models. We deliver the detection, the policy, and the signed evidence in front of them.
For the CISO
Compliance you can prove, in real time.
Track every AI request live. See what was blocked, what went to cloud, and prove it all with one click when the regulator calls.
- Live overview of all AI traffic
- 100% of calls logged and auditable
- One-click forensic export for regulators
Requests today
14,205
Blocked locally
13%
To cloud
87%
Ed25519-signed · tamper-evident · one-click forensic export
Proof of evidence
Cryptographic proof, ready for the regulator.
Every decision is hash-chained and signed. The log can't be altered without breaking the chain. When the regulator comes, you have evidence built for regulatory scrutiny, not a CSV file.
- Hash-chained, Ed25519-signed audit log
- Verifiable without contacting CareProxy
- Supports GDPR Art. 32, ISO 27001 and NIS2
Try it yourself
Type a request. Watch the decision.
A simplified version of CareProxy's triage that runs in your browser. Type or pick a prompt and watch how it is classified and routed. Nothing leaves your browser.
Detected signals
- PIIICD-10 diagnosis code
Patient data → your on-premise model in your own network.
Logging is not evidence
Your gateway logs. But can it prove?
What a gateway log shows
- Request ID, timestamp and which model was called
- Token usage and response times
- Raw traffic, depending on configuration
- Rows that whoever operates the system can edit
- An engineering artefact, built for debugging
What the regulator asks
- Which personal data did the AI see, and on what legal basis?
- What was removed, blocked or passed on?
- Can you prove the log is complete and unaltered?
- Can a third party verify it without trusting you?
- Does the evidence layer stand outside the system it documents?
A gateway log is an engineering artefact. Evidence is something else. CareProxy writes the latter: signed, hash-chained, and without sensitive content in the log.
Do you know exactly what your gateway log contains today? It is a ten-minute check. We will show you how.
Ready to take control of your AI infrastructure?
CareProxy is under active development. Join the waitlist to be notified as soon as pilot installations open.
Or reach out directly kontakt@careproxy.dk