Short answer: Jev can be used safely by a UK business, but not for everything, and not without a bit of paperwork. TypeSafe hosts it in the United States, offers a proper data processing agreement, says it won't train on your data - and gives no uptime guarantee. That combination is fine for some jobs and wrong for others. This guide explains which, in plain English.
We read TypeSafe's privacy policy, Data Processing Addendum and customer agreement so you don't have to. Everything below reflects those documents at the time of writing. They will change as Jev leaves early access, so check the current versions before you sign anything.
First, What Is Jev?#
Jev is a model from TypeSafe AI, released in early access on 15 September 2026. Unlike ChatGPT or Claude, it doesn't write text. You send it some information (TypeSafe calls this the state) and a typed question, and it answers with one of three things:
- Choice - pick one option from a list ("billing", "technical", "sales")
- Score - rate something on a scale you define ("low", "medium", "high" urgency)
- Noul - the probability that a statement is true ("this email is spam")
Every answer comes with a probability and a confidence score. TypeSafe says Jev answers in 70-500 milliseconds and charges $0.042 per million input tokens, with output free. Those are the company's own figures from its launch - treat them as best-case until you've tested your own workload.
So Jev is built for the small decisions inside software: sorting, routing, scoring, checking. That matters for safety, because what you send to Jev is the main thing that decides your risk.
Where Does Your Data Go?#
TypeSafe's privacy policy states that its services are hosted in the United States. So every piece of text you send to Jev leaves the UK.
Under UK GDPR, sending personal data to the US is a restricted transfer. It's allowed, but only with a legal mechanism in place. There are two routes:
- The UK-US "data bridge" - works only if the US company has self-certified to the Data Privacy Framework and opted into the UK Extension. You can check any company on the official Data Privacy Framework list. We couldn't confirm TypeSafe is on it at the time of writing.
- Contract clauses - TypeSafe's Data Processing Addendum includes the EU Standard Contractual Clauses (the controller-to-processor module) and the UK Addendum. This is the route most UK customers will rely on.
If you use route 2, the ICO expects you to carry out a transfer risk assessment - a short document weighing up whether the data is adequately protected once it reaches the US. For low-risk data this can be brief. For sensitive data it's a real piece of work.
The practical shortcut: if the text you send contains no personal data, most of this doesn't apply. A lot of Jev's best use cases - categorising product feedback with names stripped out, reranking search passages, checking another model's output - can be designed that way.
Will TypeSafe Train on Your Data?#
TypeSafe's privacy policy says it will not train or fine-tune any AI models on your prompts or other input. Its customer agreement adds that customer data won't go into a training dataset without your prior consent.
Two things to note:
- The customer agreement still lets TypeSafe use customer data for telemetry and to improve its services. "Not used for training" is not the same as "never looked at".
- Zero data retention is for enterprise customers only. On standard terms, the privacy policy says data is kept "for as long as reasonably necessary", with no fixed period.
If your data protection officer needs a retention period in days, you'll need an enterprise conversation with TypeSafe - or keep personal data out of what you send.
The Uptime Question#
This is the one most people miss. TypeSafe's customer agreement provides the service "as is" and "as available". There is no service level agreement, no uptime percentage and no service credits if it goes down. For support, TypeSafe commits only to "commercially reasonable efforts".
The agreement also caps TypeSafe's total liability at the greater of what you paid in the previous 12 months or $50. It's governed by California law.
None of this is unusual for a product in early access. But it means Jev should never be a single point of failure. If a Jev outage would stop your support desk from routing tickets or your website from accepting enquiries, you need a fallback: a second model (or a simple rule) that takes over when Jev is slow or unavailable.
Automated Decisions: The UK GDPR Rule to Watch#
Jev makes decisions - that's its whole purpose. UK GDPR has specific rules on decisions made solely by automated means that have a legal or similarly significant effect on a person. Think:
- screening out job applicants
- approving or refusing credit, insurance or a tenancy
- deciding who gets access to a service
If you use Jev (or any AI) for decisions like these, you need safeguards: telling people, letting them get a human to review the decision, and being able to explain it. Jev's confidence scores actually help here - you can send every low-confidence decision to a person. But the legal duty is yours, not TypeSafe's.
Routing a support ticket to the billing team isn't a significant decision about a person. Rejecting a CV is.
So, Is Jev Safe? A Simple Traffic-Light Guide#
| Use case | Verdict | Why |
|---|---|---|
| Sorting anonymised feedback, tickets or documents into categories | Go | No personal data leaves the UK; a wrong label is low-harm |
| Checking another AI's answers or reranking search results | Go | Usually no personal data; easy to add a fallback |
| Scoring website enquiries for spam or urgency | Care | Contains names and emails - sign the DPA, do a transfer risk assessment, strip what you can |
| Triage of customer support messages with personal details | Care | Same as above, plus a fallback so an outage doesn't stop your queue |
| Anything involving health, criminal or other special category data | Stop | Higher legal bar for US transfers; get specialist advice first |
| Solely automated decisions about people (hiring, credit, access) | Stop | Needs human review and explanation rights by law |
| Anything that must be up 24/7 with no fallback | Stop | No uptime guarantee in current terms |
A Checklist Before You Switch On Jev#
- List what you'll send. Write down the exact fields in the state. Is any of it personal data?
- Strip or pseudonymise anything you don't need. Jev rarely needs a name to decide whether a message is about billing.
- Sign TypeSafe's Data Processing Addendum if any personal data remains.
- Check the Data Privacy Framework list; if TypeSafe isn't on it with the UK Extension, write a short transfer risk assessment.
- Update your privacy notice and record of processing to mention the new processor and the US transfer.
- Build a fallback for outages and for low-confidence answers.
- Keep humans in the loop for any decision that significantly affects a person.
Most of this is a day's work for a low-risk use case. The fallback is the part teams skip - and the part that hurts when something goes wrong.
Frequently Asked Questions#
Is Jev GDPR compliant?#
No tool is "GDPR compliant" on its own - compliance depends on how you use it. TypeSafe provides the building blocks: a Data Processing Addendum with the EU Standard Contractual Clauses and the UK Addendum, 72-hour breach notification, and a promise not to train on your inputs. You still need to do your part: lawful basis, transfer risk assessment, privacy notice and data minimisation.
Where is Jev hosted?#
In the United States, according to TypeSafe's privacy policy. There's no UK or EU hosting option in the published terms at the time of writing.
Does Jev store my data?#
On standard terms, TypeSafe keeps data for as long as reasonably necessary, without a fixed period. Zero data retention is offered to enterprise customers.
What happens if Jev goes down?#
Nothing in the contract protects you: there's no uptime guarantee and liability is capped at the greater of 12 months' fees or $50. Design your system so another model or a simple rule takes over automatically.
Can I use Jev for recruitment screening?#
You can use it to help, but not to make solely automated rejections without safeguards. UK GDPR requires human review for decisions with significant effects on people. A sensible pattern is letting Jev sort CVs into "clearly relevant" and "needs a look", with a person making every rejection.
Jev is genuinely interesting: a fast, cheap way to make the small decisions that currently cost a full language model call. The safety questions are answerable - they just need answering before you switch it on, not after.
If you want to know whether it's worth it for your product, our Jev Switch Check tests your real examples against Jev and a UK-hostable backup, and tells you what you'd save, what accuracy you'd keep, and which calls should stay where they are. Send us a sample and get a free savings estimate.
AI engineer at BrightBit Digital
