Data processing agreement
Version 5.3, 17 November 2026. This agreement exists in Dutch (the "verwerkersovereenkomst") and in English, with the same content. If the two texts differ, the Dutch text prevails.
This agreement follows the structure and the minimum content of the European Commission's standard clauses for controllers and processors (Implementing Decision (EU) 2021/915), and contains everything Article 28 GDPR requires. It is not those clauses word for word.
1. Who this agreement is between
- The customer: the organisation that has a ScamCanary business account, or a free pilot under a signed pilot letter. For the personal data of its people, the customer is the controller: it decides that its people are trained, who, and with which settings.
- ScamCanary: scamcanary VOF, a Dutch general partnership (vennootschap onder firma) of Job van Breukelen and Oliver Milner, at Amsterdamsestraatweg 152BIS, 3551 CM Utrecht, the Netherlands, registered with the Dutch Chamber of Commerce (KvK) under number 42224323. ScamCanary is the processor: it processes that personal data only on the customer's behalf, to provide the service.
Another company in the same ScamCanary group (a "Multiple companies" account) may join this agreement as a controller by being added to Annex 1.
In this agreement, "subscription" includes a free trial and a free pilot.
How this agreement fits with the others. This agreement, with its annexes, is part of the ScamCanary terms of service. Where they differ:
- on personal data, this agreement prevails over the terms of service and over a pilot letter;
- the liability limit in a signed pilot letter applies to the pilot it covers (section 14);
- on everything else, the terms of service apply, and for a pilot the pilot letter.
Transfer. Under section 13 of the terms of service, ScamCanary may transfer its agreement with the customer to a private limited company (B.V.) it sets up to continue ScamCanary and in which the partners together hold most of the shares, or to the partner of scamcanary VOF who continues ScamCanary. It tells the customer by email at least 30 days before. This agreement then passes with it, and that company or partner becomes the processor. If the customer does not want the transfer, it may end its subscription before the transfer takes effect, and is refunded for the unused period.
2. What is processed, why, and for how long
Annex 1 describes the processing: whose data, which data, what is done with it, why, and how long it is kept. The purpose is security awareness training for the customer's people. ScamCanary does not use the customer's personal data for anything else, apart from the limited purposes of its own in section 18. It does not sell it, and does not use it to train AI models.
This agreement takes effect when the customer accepts it (section 20). It applies for as long as the customer uses ScamCanary, and afterwards until ScamCanary has deleted the customer's personal data (section 11). It ends by itself once that deletion is done, and cannot be ended on its own while ScamCanary still processes that data.
3. Instructions
- ScamCanary processes the personal data only on the customer's documented instructions. Those instructions are:
- this agreement;
- the settings the customer chooses in its ScamCanary account: who is invited, the practice-email levels and schedule, the training plan, whether insights are shown per team or per person, the leaderboard, and links between people and sister companies;
- the actions its admins take in the portal, such as removing a person, turning a reported email they confirmed as phishing into a practice exercise, asking ScamCanary to report a look-alike domain, or uploading a policy to make a training module from it;
- that a reported real email stays private to the person who reported it until they share it with the admins (section 13);
- anything else agreed in writing (email is enough).
- The exception is where EU or Member State law requires ScamCanary to process the data. ScamCanary then tells the customer first, unless that law forbids it.
- If ScamCanary thinks an instruction breaks the GDPR or other data protection law, it tells the customer straight away, and need not follow it until the customer confirms or changes it.
- An instruction that needs work beyond what the product does, and beyond ScamCanary's own duties under this agreement and Article 28 GDPR, may be charged at reasonable cost, agreed with the customer in advance.
4. Confidentiality
Only people who need access to provide, maintain or support the service can reach the personal data. Today that is the two partners of scamcanary VOF, Job van Breukelen and Oliver Milner. Each has signed a written confidentiality statement, and anyone given access later signs one before they get it. The duty continues after someone leaves. Both signed theirs on 7 November 2026.
5. Security
ScamCanary takes at least the technical and organisational measures in Annex 2, to protect the personal data against loss, destruction, alteration and unauthorised access or disclosure. It may improve them over time, but never lower the overall level of protection.
6. Sub-processors
- The customer gives general authorisation for the sub-processors in Annex 3.
- ScamCanary tells the customer at least 30 days before adding or replacing a sub-processor. It sends an email to the account owner, and updates the list in its privacy policy (section 5, "Who helps us", at scamcanary.com/en/privacy). This also applies when ScamCanary moves a task to another AI provider (section 8 of the terms of service).
- The customer may object on reasonable data protection grounds within those 30 days. If ScamCanary cannot resolve the objection, the customer may end its subscription before the change takes effect, and is refunded for the unused period.
- ScamCanary engages each sub-processor under a written contract with, in substance, the same data protection obligations as this agreement.
- ScamCanary remains fully liable to the customer for the performance of its sub-processors' obligations, within the limits of section 14, and tells the customer if one fails to meet its obligations.
- On request, ScamCanary shares a copy of a sub-processor's data processing terms. Most are public, and Annex 3 links to them.
7. Transfers outside the European Economic Area
- The customer's personal data is stored in the European Union: Railway's EU West region (europe-west4), Amsterdam.
- Some sub-processors are American companies. Where a sub-processor processes personal data outside the EEA, or can access it from there, the transfer relies on a mechanism in Chapter V GDPR, listed per sub-processor in Annex 3:
- the EU-US Data Privacy Framework (an adequacy decision), where the sub-processor is certified under it;
- otherwise, or as well, the Commission's standard contractual clauses (Decision (EU) 2021/914), in the module that fits ScamCanary's role as processor (Module 3, processor to processor).
- ScamCanary has assessed these transfers, and shares that assessment on request.
- ScamCanary does not transfer personal data outside the EEA in any other way without the customer's instruction.
8. Requests from the customer's people
- If someone asks ScamCanary directly to see, correct, delete, restrict, object to or take their data, ScamCanary passes the request to the customer within 5 working days. It does not answer the request itself, unless the customer asks it to.
- ScamCanary helps the customer answer such requests. Much of it can be done in the product:
- an admin can correct someone's details, pause them or remove them;
- each person can see their own results and training;
- the customer's owner can download all of the customer's data at any time (section 11).
9. Other help ScamCanary gives the customer
Taking into account what ScamCanary knows about the processing, it helps the customer with:
- security (Article 32);
- personal data breaches (Articles 33 and 34, and section 10 below);
- data protection impact assessments (Article 35) and any prior consultation with the supervisory authority (Article 36). Systematic monitoring of employees can require a DPIA. For Dutch employers, the Autoriteit Persoonsgegevens lists it among the processing that needs one. ScamCanary provides on request a description of the processing, its risks and its safeguards, for the customer's DPIA;
- keeping data accurate: ScamCanary tells the customer without delay if it learns that data is wrong or out of date.
10. Personal data breaches
- ScamCanary informs the customer without undue delay, and at the latest within 24 hours, after becoming aware of a personal data breach affecting the customer's data (Article 33(2)). ScamCanary is aware once it has a reasonable degree of certainty that a breach has happened; it investigates a suspected breach straight away, so this moment is not put off. This gives the customer time to meet its own 72 hours under Article 33. The first notice may be short; it is not held back until everything is known.
- The notice goes by email to the account owner, and to the privacy contact in Annex 1 if one is given. It contains, as far as then known:
- what happened, with the categories and approximate number of people and records affected;
- the likely consequences;
- what ScamCanary has done or proposes to do, including to limit the harm;
- a contact point for more information.
ScamCanary may give this information in phases: what it has at first, and the rest as soon as it knows more, without further undue delay.
- ScamCanary helps the customer notify the supervisory authority and, where needed, the people affected. It does not notify them itself on the customer's behalf unless the customer asks.
11. At the end
- The customer's owner can download all of the customer's data at any time, from the company profile in the portal, as a ZIP file of spreadsheets (CSV): its people, practice-email results, training records, reported emails, settings, company policies, its own rules module and activity log. The file leaves out what participants were promised stays their own: phone numbers for practice texts, mini-lesson answers, and real emails they checked but never sent to the customer.
- When the subscription ends, the customer chooses: return or deletion. Return means the customer gets this file: it stays available to the owner in the portal until the data is deleted, and ScamCanary sends it on request. Either way, ScamCanary deletes all of the customer's personal data 30 days after the end, or sooner if the customer asks, and confirms the deletion by email to the owner. The exception is where EU or Member State law requires ScamCanary to keep it.
- Encrypted backups are overwritten on a 14-day cycle, so the last copy is gone at most 14 days after the deletion. Until then, a backup is used only to restore the service, and this agreement keeps applying to it.
- Sub-processors delete their copies under their own terms (Annex 3).
12. Information and audits
- ScamCanary makes available the information needed to show it meets this agreement and Article 28. It answers the customer's questions promptly and in writing, and shares this agreement, `scamcanary.com/security` and a summary of its latest security review.
- If that is not enough, the customer may audit, itself or through an independent auditor bound to confidentiality. It starts with a written questionnaire, which ScamCanary answers promptly and in full. If the answers are not enough either, the customer may audit on site, at most once a year and announced 30 days ahead. It may also audit on site outside that rhythm, announced at least 5 working days ahead, or sooner where a supervisory authority requires it:
- after a personal data breach affecting its data;
- when a supervisory authority asks for it;
- when there are concrete signs that ScamCanary does not meet this agreement.
ScamCanary has no data centre of its own: its hosting and network providers (Annex 3) do not admit their customers' customers. For them, ScamCanary provides on request the latest independent audit reports or certifications the provider makes available, and the customer's audit covers them through those. An on-site audit takes place where ScamCanary manages the processing: at its place of business in the Netherlands or, while it has no office, where its partners manage the systems, during working hours, or by video call if both parties prefer. The auditor does not see other customers' data or ScamCanary's security keys, and keeps what it learns confidential. Each party pays its own costs.
- The results are shared with the competent supervisory authority on request.
13. What the customer is responsible for
The customer is responsible for:
- having a lawful basis for the training. The customer relies on its legitimate interest (Article 6(1)(f) GDPR) and makes its own balancing test. Its people still choose to take part: each one accepts an invitation before any practice email, and can stop at any time in their settings, which is how they exercise their right to object. Taking part is not consent in the sense of the GDPR, and neither party relies on it as such;
- informing its people beforehand about why and how they will practise;
- not using anyone's results for appraisal, sanctions or any other decision about their employment. Its owners and admins see per-person results (there is no line-manager role), and each time an admin opens a person's results page it is logged (Annex 2, "Logging");
- what its admins write in a free-text field that ScamCanary sends outside ScamCanary on their request, such as the note in a look-alike report (Annex 1): the customer makes sure its admins put no personal data there.
Before the first practice email. The customer warrants that, before it invites anyone to take part:
- where the law requires it, its works council has approved the training (in the Netherlands: Article 27(1)(l) of the Works Councils Act, WOR, and sub k for the processing of personnel data); and
- it has carried out a data protection impact assessment (DPIA) where one is required (section 9).
The customer confirms this when it accepts this agreement (section 20). ScamCanary does not check it, and relies on that confirmation.
Reported real emails. The customer instructs ScamCanary to keep a reported real email private to the person who reported it, until that person shares it with the admins. The customer already has the original in its own mailbox, so ScamCanary gives it no other way to see an unshared report.
14. Liability
ScamCanary's liability under this agreement and under the terms of service is one liability, not two: section 9 of the terms limits both together. That section sets a separate, higher limit for loss caused by a breach of this agreement or of data protection law, excludes indirect loss, and makes a claim lapse unless it is made in writing within 12 months after the loss was discovered. For a free pilot, the limit in the signed pilot letter applies instead. These limits do not apply to loss caused by intent or deliberate recklessness, or to liability the law does not allow to be limited. Nothing in this agreement or in those limits restricts the rights that a data subject or a supervisory authority has under the GDPR. Where ScamCanary and the customer are both liable to a data subject, each bears the share of the compensation that matches its part in the damage (Article 82(5)). Section 9 of the terms also covers the customer's responsibility for its own duties under section 13, including claims against ScamCanary from its people or a supervisory authority.
15. Suspension and termination
- If ScamCanary breaches this agreement, the customer may tell it to stop processing until it complies again.
- The customer may end the subscription with immediate effect, and is refunded for the unused period, if ScamCanary:
- does not comply again within one month of such a suspension;
- is in serious or persistent breach of this agreement or the GDPR;
- does not follow a binding decision of a court or supervisory authority about its obligations.
- ScamCanary may end the subscription if the customer insists on an instruction that ScamCanary has said breaks the law (section 3.3).
- However the subscription ends, section 11 applies, and this agreement ends as section 2 says.
16. Changes to this agreement
ScamCanary may update this agreement to follow a change in the law or in a regulator's guidance, or to update the annexes. It announces a change that reduces the customer's protection 30 days in advance. The customer may then end its subscription and be refunded for the unused period. Changes to Annex 3 follow section 6. ScamCanary keeps every version of this agreement, and sends any of them on request.
17. Customers in the United Kingdom
For a customer whose processing falls under UK data protection law:
- references to the GDPR mean the UK GDPR and the Data Protection Act 2018;
- "supervisory authority" means the Information Commissioner's Office (ICO);
- the EEA in section 7 includes the United Kingdom.
The UK recognises the EEA as adequate, so the data can be stored in the EU, and no separate transfer agreement (IDTA) is needed between the customer and ScamCanary. Where a sub-processor processes UK personal data in the United States, or can access it from there, the transfer relies on the UK Extension to the Data Privacy Framework, or on the UK Addendum (the International Data Transfer Addendum to the standard contractual clauses), as Annex 3 lists per sub-processor.
The ICO's guidance on monitoring workers expects a DPIA before employees are monitored; section 9 applies to it in the same way.
18. What is outside this agreement
Some processing is not done on the customer's behalf. ScamCanary does it as controller, for its own limited purposes, and the privacy policy describes it:
- keeping accounts secure: the sign-in and security log of every account, 365 days;
- preventing a free trial being taken twice: which addresses and company email domains received practice emails during a free trial, stored only as a keyed hash that cannot be read back, and kept two years after the last practice email;
- counting website visits, without cookies;
- billing, sales and support contact with the customer's contacts, including ScamCanary's own records of its customers. Payments go through Stripe, which processes the customer's billing contacts, not its people's data;
- family accounts, which are not part of a business account.
19. Law and courts
Dutch law applies to this agreement. Any dispute is decided exclusively by the court in Utrecht (Rechtbank Midden-Nederland), as section 12 of the terms of service says. This does not affect a data subject's right to go to a court or a supervisory authority under the GDPR, nor, for a customer in the United Kingdom, the application of UK data protection law (section 17).
20. Accepting this agreement
- A business customer accepts this agreement in one of two ways:
- when it creates its business account and accepts the terms of service, of which this agreement is part; or
- when it signs a pilot letter that refers to this agreement.
- The person who accepts confirms that they may bind the customer, and confirms for the customer what section 13 says about the works council and the DPIA. ScamCanary keeps a record of who accepted, when, and which version: for an account, who created it and the date, with the version of this agreement that applied that day; for a pilot letter, the signed copy.
- A customer that wants a copy signed by both parties gets one, signed for scamcanary VOF by one of its partners.
- Either way, this agreement is in writing, including in electronic form, as Article 28(9) GDPR requires.
Annex 1: The processing
Parties
- Controller: the customer, as named in its account's company details or in its pilot letter: its legal name, address, and a contact (the account owner, or the customer's privacy contact or DPO if it gives one). Other companies in the same ScamCanary group that join: their name, address and contact as shown in the account, and the date they were added.
- Processor: scamcanary VOF, Amsterdamsestraatweg 152BIS, 3551 CM Utrecht, the Netherlands, KvK 42224323. Contact: Job van Breukelen and Oliver Milner, partners, [email protected].
Whose data
- The customer's employees, and anyone else the customer invites to take part, such as contractors ("participants" below).
- The customer's admins and account owner.
- People who work for a sister company in the same group, when the customer links them ("works for").
- Colleagues and contacts whose name and job title appear in practice emails.
- Anyone whose details appear in an email a participant reports as suspicious, typically the sender.
Which data
| What | Details |
|---|---|
| Who someone is at work | Name, work email address, department, job title, role in the portal, and when they were invited and joined |
| Their account | Password (stored only as a bcrypt hash), two-step secret (encrypted), language and display preferences. If someone signs in with Microsoft or Google: the email address and, for a new account made from an invitation, the name that sign-in gives, and nothing else from it |
| Personalisation details (optional) | What a participant or admin adds so practice emails fit better: a short description, city, employer, role, services used, age band |
| Practice emails | Each email ScamCanary wrote and sent them (subject, sender, text), its level, and when it was sent, clicked or reported. The text can name real colleagues and their job titles, because a realistic practice email comes from someone the participant knows. ScamCanary stores no IP address or browser details for a click. Its hosting and network providers keep ordinary access logs, which include IP addresses, for security (Annex 3) |
| Reported emails | For an email a participant reports: the sender's address, a different reply-to address if the email has one, the subject, an excerpt (the first 2,000 characters of the text), the automatic check's findings, who reported it, and whether they said they had already clicked or filled something in. Attachments are never stored |
| Training | Which modules someone started and finished, when, and which version. Answers given inside a module are not stored. Mini-lessons (short quiz cards after a report): the answer given and whether it was right. Also points, level, weekly game scores, and which monthly scam stories they were sent and opened |
| The Outlook add-in | The Microsoft sign-in (email address or username) of the person reporting, to confirm who reported |
| Microsoft 365 group (optional) | Only if the customer connects its Microsoft 365. The names of its groups, so an admin can pick one. Then, once a day, the people in the chosen group: name, work email address, job title, department, their Microsoft user ID, and whether the account is switched on. Only people who become participants are stored, with their Microsoft user ID. The sync's own app gets no access to anyone's mailbox |
| Direct delivery (optional) | Only if the customer's Microsoft 365 admin approves ScamCanary's delivery app. ScamCanary then stores the customer's Microsoft tenant ID, when it was approved, and whether Microsoft found a mailbox for each participant, and places each practice email straight into the participant's own Inbox through Microsoft Graph, instead of sending it over the internet. For that the app holds Microsoft's application permission Mail.ReadWrite. Microsoft grants it for every mailbox in the customer's organisation unless the customer limits it to the participants' mailboxes. ScamCanary recommends doing that with Exchange Online's RBAC for Applications, which replaces the older application access policy, and helps the customer set it up on request. ScamCanary uses it only to create that one message in that one Inbox: it never reads, lists, moves, sends or deletes anyone's mail, and stores nothing from the mailbox. The customer ends it by turning it off in ScamCanary or removing the app in Microsoft. Microsoft processes the placed message as the customer's own provider, not as ScamCanary's sub-processor |
| Activity log | Who in the customer's account did what, and from which IP address, for example "invited a person" or "opened a person's results" |
| Look-alike domains | Domains that look like the customer's own, found by a daily check of public DNS, with what was found (mail set up, a website, on Google's list of dangerous sites). No personal data is needed for this. Only when an admin asks ("Report it"): which admin asked, and when and to whom the report went. The admin's note is sent, not kept (see "What is done with it") |
| Company policies (optional) | A policy (PDF) the customer uploads for its people to read and accept, and who uploaded it. For each person: which version they accepted and when, and the reminders sent about it |
| Own rules module (optional, once ScamCanary offers it to the customer) | A security policy the customer uploads or pastes, or its answers to six questions, to have a training module written from it. Of an uploaded or pasted policy, only its text with identifying data removed is kept, with the module; the file itself is kept only if the customer also makes it a company policy for its people to accept. The answers to the questions are kept as typed, including the contact person the customer names for reporting, who is shown to its people in the module. The customer uploads only what the module needs: no passwords, keys or network diagrams |
| Phone number (optional) | Only if a participant adds it themselves to receive practice text messages, with their consent. Not shown to the customer, and no text messages are sent yet |
| Practice phone calls (optional, only for a customer that has them switched on) | Only if practice calls are switched on for the customer. During a call: the participant's voice and the words it turns into, which can be anything they choose to say. These pass through to make the call work and are not kept. ScamCanary stores no recording and no transcript, only counts of what the call used (seconds, tokens, characters) and whose call it was, for the spending limits. The words shown on screen stay in the participant's own browser until the page closes |
| Risk level (optional) | Only if the customer chose per-person insights: low, medium or high, worked out by fixed rules each time the page opens, and not stored |
Sensitive data. ScamCanary does not ask for special categories of personal data. A reported email may contain anything, though, including sensitive data about the sender, the reporter or others. The safeguards:
- only an excerpt is kept;
- the check is done by fixed rules, without AI;
- only after the reporter has shared it with the admins, an admin has marked it as phishing, and an admin chooses "Make a practice exercise", does any of it go to an AI (Annex 3, Mistral AI): at most 1,500 characters of its subject and excerpt, stripped of identifying data first. The To and From lines, names in greetings and sign-offs, email addresses, links, phone numbers and account numbers are removed automatically. This is pseudonymisation, not anonymisation: other names or details in the text can remain. The AI returns a short description of the kind of scam, which the admin reads and can edit. The original email is never re-sent, and the person who reported it gets no practice email based on it;
- until the reporter sends it to the admins, only the reporter can see it;
- if they never do, its subject, sender, text, findings and who reported it are emptied within an hour after their link expires: 60 minutes after a report from the Outlook button, 24 hours after a forwarded email. What remains is a row for counting, which holds no one's details except a different reply-to address if the email had one;
- everything is deleted after 90 days.
What is done with it. ScamCanary:
- stores and organises participants' details;
- writes practice emails with AI and sends them, or, when the customer approved direct delivery, places them in the participant's own Inbox. Once it is live, practice emails to a customer without direct delivery go through ScamCanary's own mail server at TransIP (Annex 3);
- records clicks and reports, and checks reported emails;
- shows lessons and training modules, and records progress;
- works out statistics for the customer's reports;
- sends invitations, reminders and notifications;
- if practice calls are switched on, runs spoken practice calls with an AI caller, which starts every call by saying that it is a scamcanary practice call and that the voice is an AI, and shows the participant afterwards what gave the call away. If the AI thinks the participant agreed to what the caller asked, it asks them whether they did. That guess is shown to the participant alone and is not stored, and only their own answer counts;
- when the customer connects its Microsoft 365, keeps the participants in step with the group it chose: it invites people who join the group and pauses people who leave it, or whose account is switched off;
- checks every day for look-alike domains of the customer's own domain. Only when an admin asks, it reports one ("Report it"): an email to the registrar's and the hosting company's abuse desks (for .nl domains, to SIDN), and the website to Netcraft. The report goes out in ScamCanary's name, and replies come to ScamCanary's support address. It contains the domain, the customer's company name, the technical evidence and, if the admin adds one, a note, and no person's details from ScamCanary's records. The note is the customer's instruction: ScamCanary sends it as the admin wrote it, with links removed, and does not keep it. The customer makes sure its admins put no personal data in it (section 13). The registrars, hosting companies, SIDN and Netcraft handle the report as independent controllers for their own abuse handling, not as ScamCanary's sub-processors. Netcraft is in the United Kingdom, which has an EU adequacy decision until 27 December 2031;
- if the customer uses it, writes a training module from the customer's own policy with AI (Annex 3, Mistral AI): the policy text or the answers to the questions, stripped of identifying data in the same way as a reported email, go to the AI; the contact person named in the answers never does. The customer reviews every rule before publishing. The module and its text are used only for that customer, and never reused in practice emails, modules or templates for others.
When the customer links a person to a sister company in its group, that company sees the person's results too.
How long it is kept
| What | Kept until |
|---|---|
| Participant details, practice-email history and training records | The customer removes the participant, or the subscription ends (section 11) |
| The voice and words of a practice phone call | Not kept: gone when the call ends |
| Personalisation details | Cleared 90 days after they were last set |
| Reported emails, and a practice exercise made from one | Deleted after 90 days. Of a report the reporter never sent to the admins, the content and who reported it are emptied within an hour after its link expires |
| Activity log | 365 days |
| Usage counts (who used ScamCanary AI, a practice exercise, the own rules module or a practice call, when and how much; never the content) and ScamCanary AI summaries (totals only) | 365 days |
| Look-alike domains, and who asked to report one | The subscription ends (section 11) |
| Company policies, own policy text and the module made from it | The subscription ends (section 11). A person's acceptances go when the person is removed. The portal can archive a policy, but has no way to delete a policy or the module earlier; ScamCanary does that on the customer's request by email, within 30 days, and confirms it |
| Team spreadsheets uploaded but not confirmed | Deleted after one day |
| Sign-in sessions and one-time links | Deleted when they expire |
| Backups | Encrypted, one a day, the last 14 kept: on the hosting provider's storage, and a copy outside it (Annex 3, Cloudflare) |
A job in the product applies these deletions every hour.
Annex 2: Technical and organisational measures
These are the measures in place on 11 November 2026. `scamcanary.com/security` describes them in more detail.
Encryption
- All traffic uses HTTPS (TLS), and browsers are told to use nothing else (HSTS, two years).
- Passwords are stored only as bcrypt hashes. Sign-in sessions, password-reset links and verification links are stored only as SHA-256 hashes.
- Two-step secrets and add-in keys are encrypted with AES-256-GCM, with a key stretched by scrypt.
- Backups are encrypted with AES-256-GCM before they are written.
- An encrypted copy of each backup is kept outside the hosting provider, in the EU. Once a week the newest copy is downloaded, decrypted and checked, to prove the service can be restored from it.
Access
- Every page and every action checks that the person may see or change that customer's data. An admin sees only their own company, plus the sister companies they run in the same group. Only the customer's owners and admins see per-person results; there is no line-manager role.
- Only ScamCanary's two founders can reach the founders' area, and only with two-step verification on and a confirmed email address.
- Passwords are at least 12 characters and are checked against common weak passwords. Repeated failed sign-ins lock the account for a while.
- Two-step verification is available to every user.
- The Outlook add-in, the Microsoft 365 sync and direct delivery each run through their own Microsoft app, so approving one never grants another's permissions. ScamCanary believes which Microsoft organisation approved an app only from a sign-in Microsoft signed, and one Microsoft organisation can be attached to one customer only.
Protecting the website
- A strict content security policy, so injected code cannot run.
- Rate limits on sign-in, sign-up, invitations and the report button.
- Cloudflare's network protection in front of the site.
- Practice emails never contain a real outside web address, email address, phone number or bank account number, and AI-written text that does is rejected.
Keeping less
- A click on a practice email stores no IP address or browser details in ScamCanary. Railway and Cloudflare keep ordinary access logs, which include IP addresses, under their own terms, to run and protect their networks.
- The AI that writes practice emails receives only what an email needs: the participant's first name, job title and department, the company's name and sector, the optional personalisation details, the names, job titles and departments of up to twelve colleagues it may mention or use as a sender, and, in a group, the names, job titles and company of up to eight contacts at a sister company. It never receives email addresses, passwords, results or anyone's email history.
- Requests to Mistral go to its EU endpoint and are processed in the EU/EFTA. ScamCanary's account is opted out of training, and Mistral deletes them after at most 30 days.
- ScamCanary AI, the summary for admins, receives totals for the company and for departments of 3 or more people, with the departments renamed; a smaller department counts only in the company total, and the advice never names it. It also receives the admin's question with email addresses, phone numbers and web addresses removed. It receives no names from ScamCanary's records. A question an admin types can still contain a name; asked about a person, ScamCanary AI declines and points to the rule-based risk levels. It never assesses a person.
- ScamCanary gives LiveKit and Mistral no name or email address for a practice phone call: the call is known by an internal number. What the participant says themselves does reach them. Recording, transcripts and session analysis are off at LiveKit, and the caller's logs hold errors and counts only. Before a call's words go to the AI for the after-call question, email addresses, web addresses and long numbers are removed, also when they were said aloud.
- The retention periods in Annex 1 are applied automatically, every hour.
Availability and recovery
- An encrypted snapshot of the database is made every day, and the last 14 are kept.
- An encrypted copy of each daily snapshot is kept outside the hosting provider, in Cloudflare's EU storage, and a restore of the newest copy is tested every week.
- The founders are alerted by email when a part of the service fails.
Logging
- Each customer account has an activity log of what admins do, kept 365 days. This includes each time an admin opens a person's results page (once an hour per admin and person, so a reload does not fill the log).
- Every user account has a security log (sign-ins, failed sign-ins, lockouts, password and two-step changes), kept 365 days. ScamCanary's founders can see it; the customer cannot.
Governance
- The platform had a full security review in September 2026, with further rounds in October 2026, and its findings are tracked.
- Software dependencies are checked for known vulnerabilities (GitHub Dependabot).
- Changes are made on a separate branch, and reach the live platform only when a founder merges them.
- Access to production systems is limited to the two founders.
Helping the customer
- Admins can correct someone's details, pause them or remove them in the portal. Removing someone deletes at once their practice emails, personalisation details, mini-lesson answers and policy acceptances, and their name on reports they made. Their login goes too, with their training records and game scores, unless they still use it for another company or a family account.
- Each participant can see their own results and training.
- The owner can download all of the customer's data, as spreadsheets, from the company profile. ScamCanary makes the same file on request.
Annex 3: Sub-processors
| Sub-processor | What it does for ScamCanary | Where | Transfer mechanism | Its terms |
|---|---|---|---|---|
| Railway Corporation, 548 Market St PMB 68956, San Francisco, CA 94104, USA | Hosts the application, its database and backups | Stored in EU West (europe-west4), Amsterdam. Railway's operations are in the United States, so its staff may access data from there | Standard contractual clauses, Module 3, governed by Irish law, plus the UK Addendum (in its DPA). Railway's DPA does not say Railway is certified under the Data Privacy Framework, so these clauses are the transfer mechanism | railway.com/legal/dpa, Signed by the account holder, Job van Breukelen, and countersigned by Railway on 29 September 2026 (DocuSign envelope B9267A68). To be signed again in the name of scamcanary VOF once it is registered with the KvK; until then Job van Breukelen holds this contract for the partnership |
| Cloudflare, Inc., 101 Townsend St, San Francisco, CA 94107, USA | Protects the website and its traffic; receives reported emails forwarded to ScamCanary's report address; stores the encrypted copies of ScamCanary's backups (R2) | Cloudflare's global network. Backup copies are stored in its EU jurisdiction | Data Privacy Framework, including the UK Extension; standard contractual clauses, Modules 2 and 3, with the UK Addendum (in its DPA) | Customer DPA |
| Resend (Plus Five Five, Inc.), 2261 Market Street #5039, San Francisco, CA 94114, USA | Sends invitations, reminders and notifications, and the abuse reports an admin asks for (Annex 1, look-alike domains). Until ScamCanary's own mail server at TransIP is live, also practice emails for a customer that has not approved direct delivery (Annex 1) | Sent from its EU region (Ireland) | Data Privacy Framework and UK Extension; standard contractual clauses, Module 3 | resend.com/legal/dpa. It deletes customer data within 90 days of an account ending; while the account is active it keeps email content and logs for 30 days, and its backups for 7 |
| Mistral AI SAS, 15 rue des Halles, 75001 Paris, France | Writes the text of practice emails (see Annex 2, "Keeping less"). For ScamCanary AI: totals and the admin's question. For a practice exercise an admin starts from a reported email: at most 1,500 characters of its subject and excerpt, stripped of identifying data (Annex 1, "Sensitive data"). For a module from the customer's own policy, if the customer uses it: the policy text, stripped of identifying data in the same way (Annex 1). For a practice phone call, if switched on: plays the AI caller, so it receives the words of the call as it happens, in order to answer, straight from the caller agent and not through LiveKit; and afterwards, the text of the call for the after-call question, with email addresses, web addresses and long numbers removed (Annex 2, "Keeping less") | The EU: every request goes to Mistral's EU endpoint, which processes in data centres in the EU and EFTA | Within the EU/EFTA. For any sub-processor of its own outside the EEA, Mistral's DPA uses adequacy decisions or standard contractual clauses | Its DPA (effective 27 July 2026), accepted through its commercial terms by the account holder, Job van Breukelen, who holds this contract for the partnership. ScamCanary's account is opted out of the use of its data for training Mistral's models. Mistral keeps requests and answers for at most 30 days to detect abuse, and then deletes them. Its sub-processors |
| LiveKit Incorporated, 455 Market St, Ste 1940, PMB 195903, San Francisco, CA 94105-2448, USA | Only for practice phone calls, for a customer that has them switched on (for a customer that accepted an earlier version of this agreement without LiveKit, only after the notice in section 6). Carries the participant's voice, hosts the AI caller, and turns speech into text and text into speech. That last part is done by Deepgram, LiveKit's own sub-processor, through LiveKit Inference | Project data region EU (Frankfurt), caller hosted in Frankfurt (eu-central), speech recognition and the spoken voice on Deepgram's EU endpoints (Nova-3 and Aura-2 have one, and the project's inference region restriction sends requests only there; LiveKit's DPA says these settings decide where the speech provider processes). LiveKit's sub-processor list gives Deepgram's home country, the United States. Voice takes the shortest route, so it can pass through servers outside the EU without being stored there. LiveKit is an American company, so its staff may access data from the United States | Data Privacy Framework, including the UK Extension (its privacy policy says it is certified); standard contractual clauses, Module 3, governed by Irish law, plus the UK Addendum (in its DPA) | Its DPA, accepted through its terms of service by the account holder, Oliver Milner, who holds this contract for the partnership. To be accepted again in the name of scamcanary VOF once it is registered with the KvK. LiveKit says neither it nor its model providers log, store or train on what is sent through LiveKit Inference. Its sub-processors |
| Microsoft Ireland Operations Limited, One Microsoft Place, South County Business Park, Leopardstown, Dublin 18, Ireland | Hosts ScamCanary's mailbox for reports forwarded to [email protected] | The EU: ScamCanary's Microsoft 365 tenant is in Microsoft's European region | Within the EU. Its DPA covers any access from outside the EU with the Data Privacy Framework and standard contractual clauses | Microsoft Products and Services DPA |
| TransIP B.V., Schipholweg 9B, 2316 XB Leiden, the Netherlands | Planned, not live yet. It starts only after the 30 days' notice in section 6. Hosts ScamCanary's own mail server, which sends practice emails to a customer that is not on Microsoft 365 direct delivery. The server keeps no mail; its logs (sender, recipient address, time) are kept 14 days | Amsterdam, the Netherlands | Within the EU | The processing agreement in TransIP's general terms, accepted when the server was ordered on 29 October 2026 |
Not sub-processors. Microsoft, for direct delivery into the customer's own tenant, acts as the customer's own provider (Annex 1). The registrars, hosting companies, SIDN and Netcraft that receive a look-alike report are independent controllers for their own abuse handling (Annex 1). Stripe, for billing, is outside this agreement (section 18).