This policy describes how CyberWatch AI handles personal data in its security console for organizations, and in the personal scanner, browser extension and newsletter that accompany it.
CyberWatch AI provides a security console for organizations. It shows an organization where it stands on security, with particular attention to what its people do when a convincing message arrives, and helps it improve that position through phishing simulations, training, readiness assessments, the handling of threats its people report, remediation tasks, reporting and an artificial intelligence advisor. Alongside the console, CyberWatch AI offers a personal scanner for individuals, a browser extension that brings the scanner into the right-click menu, and a monthly newsletter.
Where an organization subscribes to the service and invites its own people into it, that organization is the controller of the records held inside its account and CyberWatch AI acts as its processor. CyberWatch AI is the controller of the remaining personal data described in this policy, such as the accounts of individuals using the personal scanner, enquiries made through the website and the newsletter list. Section 4 explains where the line falls, because it determines who should be approached about a particular record.
Questions about this policy, or about data held by CyberWatch AI, may be sent to info@cyberwatchai.com.
This policy applies to the website at cyberwatchai.com and the forms it carries, to the organization console used by administrators and employees, to the personal scanner, to the CyberWatch AI extension for Chrome, and to email sent by the service, including invitations, simulation messages, notifications, account and billing messages, and the newsletter.
It does not extend to other companies' websites or services reached through links in the service. From the moment a link is followed out of CyberWatch AI, the policy of the destination governs.
Most personal data in the service arrives when an organization sets itself up. An administrator supplies the organization's name and the names, work email addresses, roles and departments of the people being invited, and each person who accepts an invitation supplies their own name and sets a password. A password is stored only as a cryptographic hash, which means it cannot be read or recovered by anyone at CyberWatch AI. Where a person turns on two-factor sign-in, the database provider holds the authenticator secret on which it depends. Codes sent to confirm an email address are stored only as hashes and expire.
Further data is created as people use the console. It includes the outcome of each simulated phishing message, meaning whether it was delivered, opened, clicked or reported and at what time; which training modules each person completed, when, and with what result; the answers, scores and timings of readiness assessments; the organization's own answers to its security baseline questions; the risk score, label and short summary of each check a member runs on suspicious content; threats members choose to report, with the reported content, any screenshot and the identity of the person who reported it; remediation tasks and the people assigned to them; messages exchanged within the console; and the security score of the organization and a risk score for each member, recorded periodically so that movement over time can be shown.
The personal scanner holds, for each individual account, the result of each check, meaning its type, risk score, label and a short summary, but not the link, message or screenshot that was checked, together with a count of checks performed that day and the date of the most recent one. Checking is free and has no daily allowance; the count serves only to hold back automated abuse of the service. Section 5 deals with submitted content in detail.
Payment data is limited. Payments are handled in their entirety by the payment provider, and CyberWatch AI records only the confirmation of payment, the plan and billing cycle bought, and the provider's reference for the paying customer and the subscription, which is how a subscription being replaced is found and cancelled.
The website collects what a person chooses to send through its forms. Requesting a demonstration or a complimentary assessment supplies a name, an email address, an organization name, its approximate size, its industry, whether security training is currently in place and any notes added, and subscribing to the newsletter supplies an email address. Anything written to support, or in reply to a notification, is also held. A small amount of technical data arrives automatically: the hosting provider records the originating IP address, browser type and version, device type and the pages requested, in server logs kept as part of delivering the site and protecting it from abuse.
What a service declines to collect is as telling as what it gathers. CyberWatch AI operates no advertising network, no third-party analytics and no tracking pixels. It does not read, alter or record the web pages a person visits, and this holds for the browser extension as firmly as for the site. Card numbers are never requested or stored. Special category data such as health information, religious or political views and biometric data is not collected, and should not be placed into a submission, a report or a message. Personal data is not purchased from data brokers, and it is not sold to anyone.
Where an employer subscribes to CyberWatch AI and issues an invitation, that employer is the controller of the records within its organization and CyberWatch AI acts as its processor. Questions about why simulations are conducted, what is done with results and scores, or how long a record is retained are therefore properly addressed to the employer in the first instance, and CyberWatch AI will assist the employer in answering them.
The extent of an administrator's visibility is a deliberate design decision rather than an artefact of the software. An administrator can see the organization's security score and its components, figures for each department, the invitations issued and accepted, the result of each simulation for each person in terms of delivery, opening, clicking and reporting, training completions and results, readiness assessment results, each member's risk score, remediation tasks, and every threat a member has chosen to report, including the reported content and any screenshot. An administrator can also read the messages in every channel of the organization, including department channels, but not direct conversations between two people.
An administrator cannot see the content of checks a member ran without reporting them, which contribute to the organization's figures only as counts and risk levels; anything checked with the personal scanner under a personal account; the content of direct messages exchanged between two people within the console; or any password. The console states this before a check is run. A check that is not reported remains between the person who ran it and the model, and CyberWatch AI regards that undertaking as part of the product rather than a setting to be quietly revised.
Where access to organization data is requested from an account that has enrolled a second factor, the database refuses it unless that factor has been used in the current session.
When a link, message, email or screenshot is submitted for checking, whether by a member within the console, by an individual using the personal scanner or through the browser extension, that content travels to CyberWatch AI servers and is passed to the AI provider for analysis. Once the result has been returned the content itself is discarded, unless a member chooses to report it, as described below.
For a member of an organization, the check is recorded against that member as a risk score, a label and a short summary, which the member can see in their own history. When the member reports the item to their security team, the content itself and any screenshot are stored with the report so that administrators can act on it. For an individual using the personal scanner, the result is stored against their account in the same way, so that the history remains available. A check made without signing in is still analysed, and its result is recorded without being associated with any person.
Whatever the message contains forms part of the analysis. An email pasted in for checking carries with it any colleague's name, any customer's telephone number and any invoice detail that happens to be inside it. Such data is handled under this policy, but the most effective protection remains removing what is not needed before submitting.
Submitted content is not used to train artificial intelligence models. The AI provider processes it in order to return a result, not in order to improve its own models. Submitted content is not read as a matter of routine; access is confined to investigating a fault that has been reported or a suspected abuse of the service.
A simulation is an exercise in which an organization sends its own staff realistic but harmless messages in order to measure how they respond. Where an employer conducts such exercises, CyberWatch AI processes on that employer's instructions the work email addresses and names supplied for the people being tested, and the record of whether each message was delivered, opened, clicked and reported, together with the time of each event.
Where a simulated page invites an entry, the service records the fact that something was submitted rather than the characters that were typed. CyberWatch AI has no interest in holding a password and does not retain one.
Simulations are a measurement instrument and not a disciplinary one, and every organization using the service is asked to treat them accordingly. Whether an employer has told its staff that simulations take place, and what it does with the results, is a matter between employer and employee, and in many places is something the employer is obliged to explain in advance.
When a member works through a training module, the service records which module was completed, when, the result of its quiz and the points it earned. When a member sits a readiness assessment, the service records the scenarios presented, the answer given to each, whether it was correct and how long it took, the overall score, and when the assessment was started and finished. The organization's answers to its security baseline questions describe its practices rather than any person, and are recorded together with the identity of the administrator who gave them. In the same way, administrators may record compliance attestations, which are answers about controls such as records of processing, supplier agreements or backup testing, with an optional note and a pointer to evidence, together with the name of the administrator and the date. When an administrator runs a personal data breach check, the facts they enter, the categories and approximate number of people affected, the resulting decision on notification and the time of awareness are kept in the organization's breach register, as Article 33(5) of the GDPR requires, with the name of the administrator who made the assessment. Neither contains the personal data affected by a breach.
These records are used to show each member their own progress, to recommend what to learn next, and to produce the organization's figures. A small amount of progress through a module in which a member is partway may be kept in their browser so that they can resume it.
The console allows members of an organization to exchange messages in shared channels and in direct conversations between two people. Messages are stored so that the conversation remains available, together with who sent each one, when, any reactions and whether it has been read. Messages in an organization-wide channel are visible to everyone in the organization, messages in a department channel to the members of that department, and messages in any channel to the organization's administrators. A direct message is visible only to the two people in the conversation, and not to administrators.
Messages are not passed to the AI provider and are not read by CyberWatch AI as a matter of routine; access is confined to investigating a reported fault or a suspected abuse of the service. A member may edit or withdraw a message they sent within five minutes of sending it.
Administrators may ask the advisor, presented in the console as Ask CyberWatch AI, questions about their organization's security. To answer, the service sends the AI provider a summary of the organization's security data: the security score and its components, department figures, recent simulation results, open tasks, recent alerts described by their label and risk rather than their content, the answers to the security baseline questions and, for members whose risk score is high, their name and the counts of simulations, clicks, training and reports behind that score. It does not send the content of messages, of checks or of reported items, nor any email address.
The conversation itself is kept in the administrator's own browser, where the most recent forty messages are stored so that it can be continued, and it is not stored on CyberWatch AI servers. It can be cleared from within the console. Actions the advisor takes on request, such as creating a task, are recorded in the organization's account in the same way as if an administrator had taken them by hand.
The CyberWatch AI extension requests three permissions: to add entries to the right-click menu, and, through Chrome's activeTab and scripting permissions, to read a tab only at the moment a person asks it to. It declares no host permissions and installs no content scripts, so it cannot observe the pages a person visits, and it never alters them.
When a link or a passage of highlighted text is right-clicked and submitted for checking, the extension receives only the item selected and sends that item for analysis together with the session token. When a person chooses to report an email they have open in their webmail, by right-clicking it or from the extension's button, the extension reads that one email, once, from that one tab: its sender, subject and text. It shows what it read so the person can correct it, and sends it to their organization only when they confirm. The tab is never navigated or changed. Signing out of the extension clears the stored session.
An email can also be reported by forwarding it to the reporting address shown in the console. A forwarded email is received by Resend on CyberWatch AI's behalf, matched to the forwarder's account by their address, and handled exactly as a report made in the console; the forwarder receives a reply saying what happened to it. A forwarded email is accepted only from a member of an organization whose mail passes standard sender authentication. With a reported email, CyberWatch AI keeps what is needed to judge it: its sender, recipients, subject and date; its delivery and authentication headers; where its links lead; and the names, types, sizes and SHA-256 fingerprints of its attachments, but not the attachment files themselves. That record is visible to the organization's administrators, and the links, domains, addresses and fingerprints in it may be checked against the threat-intelligence services described in section 13.
Creating and operating an account, running the console for an organization, analysing submitted content and returning a result, and delivering the service that was requested are all necessary for the performance of the contract between CyberWatch AI and the customer. The same basis covers service email that a person needs in order to use the account, such as invitations, address confirmation, password resets, and notifications to administrators when a threat is reported.
Running simulations, training, readiness assessments, scoring and reporting for an organization is necessary for the performance of the contract with that organization, and rests additionally on the organization's legitimate interest in securing its business against attacks that target its people. The organization, as controller, determines its own lawful basis for including its people and is responsible for informing them.
Enforcing the limits attached to each plan, preventing abuse of the service, keeping the service secure and investigating faults rest on the legitimate interests of CyberWatch AI in operating a sustainable and trustworthy service, and in some respects on legal obligation. Responding to enquiries rests on the same legitimate interest. Accounting, tax and other statutory record-keeping rests on legal obligation.
The newsletter and other marketing email are sent only with consent, which may be withdrawn at any time, by the link in every newsletter or by writing, without affecting anything done before withdrawal. Where processing rests on legitimate interests, CyberWatch AI has considered whether that interest is outweighed by the rights of the individual, and an objection may be raised at any time using the contact details in section 22.
The analysis of submitted content, the advisor's answers and the service's written summaries are produced by a large language model operated on behalf of CyberWatch AI by its AI provider, and it is worth being direct about what that entails. The model errs in both directions. It may describe a harmless message as risky, and it may fail to recognise a genuine attack. A result should be treated as a well-informed second opinion rather than as a verdict. Every result is accompanied by an explanation precisely so that the reasoning can be judged rather than the number alone.
The security score and each member's risk score are calculated automatically from activity recorded in the service, such as simulation results, training completion and reported threats. They describe conduct within exercises and training, and where there is too little activity to support a figure, the console says so rather than presenting one.
No decision producing a legal or similarly significant effect is taken about any individual by automated means. CyberWatch AI neither makes nor permits its software to make automated decisions concerning employment. Where an employer relies on a score in reaching a decision about a person, that decision belongs to the employer, and the person is entitled to ask the employer to account for it.
Personal data is not sold. It is disclosed only to the providers necessary to operate the service, each engaged under a contract restricting them to acting on documented instructions.
Supabase provides the database, authentication and file storage, and therefore holds account details, authenticator secrets and the stored records of the service. Vercel provides hosting and content delivery, and its server logs carry the requests made to the site along with originating IP addresses. Groq operates the model that analyses submissions and answers the advisor, and accordingly receives the content of each submission and the organization summary described in section 9. Resend delivers email and receives the recipient address and the message content, including invitations, simulation messages and notifications; it also receives the emails people forward to the reporting address, and holds them for CyberWatch AI to collect. Groq also writes simulation emails at an administrator's request, from the administrator's description and the organization's name. Anthropic operates the model that coordinates the CyberWatch Engine, and so receives the questions administrators ask it, the organization summary described in section 9, and what the engine's agents find. To investigate a threat, the engine looks up the links, domains, IP addresses and file fingerprints involved, never message text, with registration and DNS services (rdap.org, Cloudflare and Google DNS), with public threat-intelligence sources (CISA, NIST NVD, FIRST, OpenPhish), and, where the service is configured, with abuse.ch, Google Safe Browsing, VirusTotal, AbuseIPDB and AlienVault OTX, each under its own policy. Paystack processes payments and receives the billing details entered with it; CyberWatch AI receives confirmation of payment, the final digits of a card and Paystack's references for the customer and subscription, never the full card number. Google distributes the browser extension through the Chrome Web Store and receives installation and listing data under its own policy. Where a visitor accepts the offer to read the website in another language, Google Translate receives the text of the pages viewed, and the request made from the visitor's browser, under Google's own policy; nothing is sent to it unless that offer is accepted.
Data is also disclosed to an organization where the service is used as one of its members, in the manner described in section 4. It is disclosed where the law requires it, on receipt of a valid court order or lawful request from an authority, and the affected person will be informed unless CyberWatch AI is prohibited from doing so. It may be disclosed where there is a good-faith belief that disclosure is necessary to prevent serious harm or to investigate fraud or an attack upon the service. Should CyberWatch AI be acquired or merged, data may transfer as part of that transaction, and this policy continues to apply until affected people are notified otherwise.
The database is hosted within the European Economic Area. The site is served through a global content delivery network, which means a cached copy of static pages may be held near the person requesting them. The email, payment and AI providers operate internationally and may process data outside the country in which it was collected.
Where personal data leaves the country in which it was collected, the transfer takes place under the terms of the contracts with those providers, which incorporate the standard protections required by applicable data protection law. Detail concerning any particular transfer is available on request.
An organization's records, including simulation results, training and assessment records, reports and their screenshots, tasks, messages and score histories, are retained for as long as the organization holds a trial or a subscription. Where a trial ends without a plan being chosen, or a subscription ends, the records are kept for ninety days so that the organization can resume where it left off, and are then deleted, sooner where the organization asks for it. A member removed from an organization loses access at once; the records of their past activity remain part of the organization's account under the same terms, unless the organization asks for them to be deleted.
An individual account is retained for as long as it exists, and for thirty days after deletion so that it can be restored if it was deleted in error. The results of checks made with the personal scanner are retained while the account exists, so that history remains available. The results of checks made without signing in are not tied to a person and are used for service statistics.
Demonstration and assessment requests are retained for twenty-four months from the most recent contact. A newsletter subscription is retained until it is withdrawn, after which a record of the withdrawal is kept so that no further email is sent. Billing records and invoices are retained for as long as tax and company law requires, typically six years. Server logs are held by the hosting provider for a short rolling period. Advisor conversations and the preferences described in section 18 remain in the browser until they are cleared there.
Depending on place of residence, an individual holds some or all of the following rights, and CyberWatch AI honours them for everyone regardless of location, on the view that drawing a line between people according to geography would be a curious way to run a security company.
These are the right to obtain a copy of the personal data held; to have inaccurate data corrected, although much of it can be corrected directly within the account; to request erasure, where the data is not required to be retained; to receive the data in a structured, machine-readable form; to request that processing be restricted while a dispute over accuracy or legitimacy is resolved; to object to processing founded on legitimate interests, and to direct marketing at any time; to withdraw consent where consent is the basis of processing; and to lodge a complaint with the data protection authority of the country of residence.
Any of these may be exercised by writing to info@cyberwatchai.com, and a response will follow within thirty days. Verification of identity may be required first, so that data is not handed to somebody else who has asked for it. Where the data sits within an organization's account, the request will usually be referred to that organization, since the records belong to it, and the person making the request will be told when that has happened.
Data travels over encrypted connections and is encrypted at rest by the database provider. Passwords are stored as hashes and cannot be read by anyone at CyberWatch AI. Access to an organization's records is enforced at the database level rather than only within the interface, so that one organization cannot reach another's data even if the interface were circumvented. Plan and billing details can be changed only by the server, never from a browser. Two-factor sign-in is available on every account and is recommended for administrators, and once a second factor is enrolled, organization data is not released to a session that has not used it.
Access to production data by CyberWatch AI personnel is confined to those who require it. Their administration view shows accounts, organizations, their members and counts of activity, which is what is needed to run and support the service, and requires two-factor sign-in once it is enrolled on the account in use; it does not show the content of checks, reported items or messages. Wider access is exercised only for the correction of faults and the investigation of abuse.
No service can promise perfect security and CyberWatch AI will not pretend otherwise. What can be promised is that where something goes wrong, it will be communicated promptly and truthfully.
Only what the service requires in order to function is used. Session storage keeps a signed-in person signed in, without which authentication would be required on every page. A small number of things are remembered in the browser and never leave the device unless described elsewhere in this policy: preferences such as the theme and which sections were last left open, progress through a training module partway completed, and an administrator's advisor conversation.
To decide whether to offer the website in another language, the visitor's browser language is read, and failing that the country the request comes from, as a two-letter code; neither is stored. The answer to that offer is remembered in the browser. Only if the visitor accepts is a cookie set, named googtrans, which tells Google Translate the chosen language and keeps later pages translated; choosing to show the original removes it.
Advertising cookies are not used and browsing is not shared with third-party trackers. Apart from that translation cookie, which is set only on request, no non-essential cookies are set, so the site carries no general consent banner.
The service is built for workplaces and is not directed at children. Personal data is not knowingly collected from anyone under sixteen. Where it appears that a child has supplied data, notification to the address in section 22 will result in its deletion.
An educational institution using CyberWatch AI is responsible for ensuring that any individual it enrols is old enough to use the service, and that it holds whatever permission the law requires.
Where personal data held by CyberWatch AI is exposed in a manner likely to place someone at risk, the relevant supervisory authority will be notified within the period the law prescribes, and the people affected will be told directly and in plain language what occurred, which data was involved, what has been done about it and what they should do in response. Where the data belongs to an organization's account, the organization is told first so that it can meet its own obligations.
This policy is revised as the service changes and as the law changes, and the date at the head of the page always reflects the current version. Where a revision materially affects rights or the use of personal data, notice will be given by email or within the service before it takes effect, rather than the page being altered quietly.
Correspondence concerning this policy, and requests to exercise any right described in section 16, should be sent to info@cyberwatchai.com. Marking the subject line "Data request" or "Privacy question" will direct it appropriately. A response will follow within thirty days and usually a good deal sooner.
Anyone dissatisfied with the response is entitled to complain to the data protection authority of their country of residence. CyberWatch AI would nevertheless prefer to be approached first, so that the matter can be put right.