Security and privacy

How we keep your customers' data safe

Follow a customer's message from the moment they open the chat to the moment you delete it, and see what protects it at every step. Where a promise is written into our DPA or policies, we link straight to it.

Last updated: 20 November 2026

At a glance

Conversations and search index
Stored in the EU (Ireland)
Encryption
TLS in transit, AES-256 at rest
Used to train AI
Never. It's in our DPA
GDPR
Data processing agreement built into our terms
UK regulator
Registered with the ICO, ZB810494
AI disclosure
On by default in website chat
Visitor identity
Optional: signed by your server, checked by us
1

They open the chat

Before anyone types, the chat says what it is, and can know who it's talking to.

  • It says it's an AI. In website chat, the default greeting introduces your AI assistant, the kind of disclosure the EU AI Act asks chatbots to make. You can change the wording, and add your own in other channels.
  • Your site can vouch for them. Optional: if they're signed in to your site or app, your server signs a short-lived token saying who they are, with a key only you and we know. We check it on every message, so a visitor can't simply claim to be one of your signed-in customers.
  • Verified details you can act on. Tokens last an hour at most. Actions can be set to use only verified details, and your chat history marks verified messages.
your-server.jsRuns on your server, never in the browser
const token = jwt.sign(
  { sub: user.id, aud: CHATBOT_ID,
    email: user.email, name: user.name },
  SIGNING_KEY,
  { algorithm: "HS256", expiresIn: "1h", keyid: KEY_ID }
);
// then pass the token to the chat widget
Your server signs The widget passes it on We check the signature
2

They send a message

It travels encrypted, straight to our servers in the EU.

  • Encrypted in transit. HTTPS only, enforced with HSTS, so browsers never fall back to an unencrypted connection.
  • Hosted in Ireland. Our app and database run on Heroku, in the EU.
Customer's browser Encrypted with HTTPS Resolve247 · EU (Ireland)
3

The AI reads it

Including any screenshot, photo or file they attach.

  • Built to resist prompt injection. The customer's conversation stays inside clear boundaries, and the AI is told never to take it as instructions.
  • Images don't give orders. When it reads a screenshot or photo, it's told to report the text in it, never to follow it.
  • Card numbers cut short. Full card numbers become the last four digits in image descriptions, text read from files, and handover notes.
Illustration of two example chatbot conversations: the AI answers a question about a photographed receipt, referring to the card only by its last four digits, and turns down a customer who tells it to ignore its instructions, offering a person instead.
4

The AI answers

From your knowledge, without ever learning from it.

  • Never used for training. Not by us, and not by OpenAI, whose models write the reply. OpenAI may keep a request for up to 30 days to check for abuse, then deletes it.
  • Private sources are never named or linked. The AI can still answer from a source you mark private, but visitors never see where it came from.
  • A person when it's needed. When it can't help, it offers to bring in your team.
Where it's written: DPA, section 7.2.3 →
Illustration of an example chatbot answer that names the business's public refund policy as its source, with a Contact a Human button below it.
5

It's stored

Mostly in the EU, encrypted, and seen only by people who need to.

  • Encrypted and backed up. AES-256 for the data we store, and daily backups, which are encrypted too.
  • Kept apart from other customers. Every dashboard request is checked against your organisation's access, and each chatbot's knowledge has its own search index.
  • Only the access we need. Our team gets the least access it needs to run the service, with multi-factor sign-in to production systems.
Where it's written: DPA, Schedule 2 →
Where each part lives
Conversations and settingsEU (Ireland) · Heroku
Knowledge-base search indexEU (Ireland) · Pinecone
Images and files sent in website chatEU, deleted within 30 days · Cloudflare R2
Writing replies, reading images, building the search indexWorldwide, under SCCs · OpenAI
App logsUnited States, under SCCs · Papertrail
Notification emailsWorldwide · Google Workspace

Every provider, on our Subprocessors page →

6

You delete it

Whenever you like. The moment you delete something, the AI loses access to it.

  • Delete exactly what you want. A whole chatbot with all its data, one knowledge source, or a single image or file a customer sent, along with what the AI read from it.
  • If you leave, deletion starts within 5 working days. Your knowledge content goes straight away and integration access is revoked. Copies in logs clear within 14 days, and copies in backups expire on their normal schedule, up to 12 months for operational logs.
Where it's written: Privacy policy →
Illustration of deleting a customer's image from the chat history: the confirmation asks to delete the image along with what the AI read from it, and warns that it can't be undone.

Around every step

Looked after, even when nothing's happening

Certified data centres

Our hosting providers' data centres hold ISO 27001 and SOC 2 certifications.

Told if something goes wrong

If a breach affects your data, we email you without undue delay, and always within the time the law requires.

Careful with uploads

Files customers send are limited to 10 MB, checked for type, and served back in a sandbox.

Dependencies watched

The code libraries we use are checked automatically for known vulnerabilities.

Found a security issue?

Tell us at [email protected], with enough detail for us to reproduce it. We'll confirm we've got it and keep you updated while we fix it.

  • In scope: resolve247.ai, app.resolve247.ai and our chat widget.
  • Please don't access or change other people's data, slow the service down, or test against our customers or staff.
  • Keep what you find private: please don't share it with anyone else.
  • We won't take legal action over research done in good faith under these rules.
  • We don't run a paid bug bounty.

Compliance

GDPR, in writing

We're your data processor. These documents set out exactly what that means.

Data Processing Agreement

Part of our Terms, so there's nothing to sign. Covers UK and EU GDPR, with the Standard Contractual Clauses and UK Addendum. Want it countersigned? Just ask.

Read the DPA →

Subprocessors

Every provider that handles your data, where, and how transfers are protected. We email you at least 10 days before any change.

See the list →

Privacy policy

How we handle personal data for our own purposes. Registered with the ICO under ZB810494.

Read the policy →

Cookie policy

The cookies our website uses, and how to choose which ones you allow.

Read the policy →

FAQ

Security questions, answered straight

The questions buyers and security reviewers ask us most. Something missing? Email [email protected].

Where is our data stored?

Your conversations, settings and your chatbot's knowledge-base search index are stored in the EU, in Ireland. Anything customers send in website chat, such as images and files, is kept in the EU for 30 days at most. Some processing happens outside the UK and EU: OpenAI writes the AI's replies and builds the search index, our application logs are kept in the United States, and notification emails go through Google Workspace. Our Subprocessors page lists every provider and how each transfer is protected.

Written in: Subprocessors page →

Do you use our data to train AI?

No. We don't use your data to train any AI model, ours or a subprocessor's. The AI looks your content up when it answers, and loses access the moment you delete it. OpenAI doesn't train on data sent through its API. It may keep a request for up to 30 days to check for abuse, then deletes it.

Written in: DPA, section 7.2.3 →

Is our data encrypted and backed up?

Yes. Every connection uses HTTPS, enforced with HSTS, and the data we store is encrypted with AES-256. The database is backed up daily, and the backups are encrypted and access-controlled.

Written in: DPA, Schedule 2 →

Who at Resolve247 can see our data?

Only people who need it to run the service and help you. Access is least-privilege, sign-in to production systems uses multi-factor authentication, and everyone is bound by confidentiality.

Written in: DPA, Schedule 2 →

Does the chatbot tell people it's an AI?

In website chat, yes, by default: the greeting introduces it as your AI assistant. The EU AI Act (Article 50) asks chatbots to make that clear. You can change the greeting, so keep the AI mention if you do, and add your own wording in other channels, such as email.

Can someone trick the AI into ignoring its instructions?

It's built to resist that, which is called prompt injection. The customer's conversation is kept inside clear boundaries, and the AI is told never to take it as instructions. When it reads a screenshot or photo, it's told to report the text in it, never to follow it.

Can we confirm who a visitor really is?

Yes, if you set up visitor identification. When your customers are signed in to your site or app, your server signs a short-lived token (a JWT) saying who they are, with a key only you and we know. The chatbot checks the signature on every message, so a visitor can't simply claim to be one of your signed-in customers.

Can we delete our data?

Yes, whenever you like. Delete a chatbot with all its data, one knowledge source, or a single image or file a customer sent, along with what the AI read from it. If you leave, deletion starts within 5 working days.

Written in: Privacy policy →

Do you have a GDPR data processing agreement?

Yes. It's part of our Terms of Service, so it applies automatically and there's nothing to sign. It covers UK and EU GDPR and includes the Standard Contractual Clauses with the UK Addendum. For a countersigned copy, just ask. We're also registered with the UK Information Commissioner's Office, under ZB810494.

Written in: DPA →

Are you SOC 2 or ISO 27001 certified?

Not as a company. Our hosting providers' data centres hold ISO 27001 and SOC 2 certifications, and the measures we take ourselves are set out in our DPA.

Written in: DPA, Schedule 2 →

Can we use it with health data? Is it HIPAA compliant?

No. Resolve247 isn't built for health records, payment card details or government ID numbers, and it isn't HIPAA compliant.

Written in: DPA, Schedule 1 →

Will you fill in our security questionnaire?

Yes. Email it to [email protected] and we'll answer in writing.

Written in: DPA, section 7.7 →

How do we report a vulnerability?

Email [email protected]. We'll confirm we've got it and keep you updated while we fix it. The rules for testing are in "Found a security issue?" above, and our security.txt file is at resolve247.ai/.well-known/security.txt.

Questions about security?

Ask our chatbot, email [email protected], or book a call and we'll walk you through it.

30 day free trial, no cc required!