Introducing CyberWatch AI: Security for the Moment Before Someone Clicks
CyberWatch AI measures how people respond to real attacks. What the free scanner does, what the console does, and the argument behind both.

One of the scenarios in our assessment library reads like this. At 08:47 a message arrives from your chief executive's name, sent from an address you do not recognize. "I'm in a meeting and can't talk. Please buy four gift cards and send me the codes. I'll approve the expense after." Four options follow, and the only correct one is to verify on a channel you already trust and then report it.
We put that scenario in front of employees because of what it measures. Asking someone "what is phishing?" tells you whether they have read a glossary. Putting them at 08:47 with a plausible request from a senior name tells you something closer to what you actually need to know, which is what they would do.
That distinction is the whole reason CyberWatch AI exists.
The question most security budgets never ask
Organizations can usually describe their technology in detail. The firewall, the mail filter, the antivirus, the backup schedule. Ask the same organization how its finance team would respond to a supplier bank-detail change arriving on a Tuesday afternoon, and the honest answer is almost always that nobody knows.
This is not a small gap. The FBI's complaint center tracks business email compromise as one of the costliest categories of reported cybercrime, and the mechanism is a message that persuades a person, not an exploit that defeats a control. MITRE ATT&CK lists phishing as an initial access technique precisely because it is where intrusions begin rather than where they end. The UK's National Cyber Security Centre has gone as far as publishing a blog post arguing that telling users to avoid clicking bad links still is not working, which is an unusual admission from a national agency and a correct one.
So the perimeter gets measured and the people get assumed. Most organizations treat an annual training completion record as evidence that the human layer is handled. It is not evidence. It records attendance.
Two products, one idea
CyberWatch AI is built around a single belief: human risk should be measured the way every other security control is measured, which means tested deliberately, reported honestly and improved where the test finds a gap. That belief produces two things.
A free scanner, for the moment of doubt
CyberWatch AI Scan answers one question: is this real? You paste a link, a message or a screenshot, and you get a risk score out of 100, a label, a plain-language summary, the specific warning signs it found, and what to do next. It is free, with no daily limit, and you do not need an account.
Two design choices matter more than the interface. First, it reads the content itself rather than checking it against a list of known scams, which means it works in any language and for any country, including an attack that is three days old and on nobody's blocklist yet. Second, it says when it is unsure. A confident wrong answer is worse than an honest uncertain one, so the verdict states what it could not determine. Our own documentation says plainly that the tool can miss a sophisticated scam or flag something genuine, and that you should treat a low-risk result as one input rather than permission, especially before sending money.
The content you paste is analyzed and then discarded. Only the scan type, the risk score and a short summary are kept. Nothing scanned is used to train AI models.
A console, for the organizations that cannot see
The second product is a security console built around a loop: baseline, train, test, measure. It sends realistic but harmless phishing simulations to an organization's own people, records who opened, clicked and reported, assigns short lessons to the specific people and topics the results point at, and turns all of it into one score management can follow month to month.
The second half of the console handles the attacks the simulations were practice for. When an employee gets something genuinely suspicious, they report it from their work inbox in one step, every report lands in one queue for the security team, duplicate reports of the same email are grouped as a single incident, and the people who reported it hear the outcome. That last part sounds like a courtesy. It is the thing that keeps people reporting.
What we deliberately do not do
Some of the clearest design decisions in CyberWatch AI are refusals.
- Simulations never collect credentials. A simulation measures whether somebody clicks. Nothing in one asks for a password, a card number or a code, and the code that assembles an AI-written simulation rejects any draft that tries to. There is no reason to harvest a real password to prove a point.
- No campaign sends itself. Every simulation requires an administrator's authorization before it goes out, and we recommend telling staff that testing happens before the first one lands.
- Administrators do not see private checks. If an employee scans something and does not report it, it stays theirs. The console exists for security visibility, not surveillance.
- Nobody is named for one click. Leaderboards are anonymous by default. A person who clicks gets a three-minute lesson on the exact tell they missed, not an email to the whole company.
- Nothing automated changes your organization. In the parts of the console where AI investigates reported email, every change it wants to make is queued as a proposal for an administrator to approve or decline. Content written by attackers reaches the system marked as untrusted, so an instruction hidden inside a phishing email can at most produce a suggestion a human then reads.
Got something suspicious right now? Paste the message or link into CyberWatch AI Scan for a free check before you act on it.
What this is not
CyberWatch AI is not a replacement for technical controls. Multi-factor authentication, patching, mail filtering and backups all do work that no amount of awareness can do, and anyone selling human-layer security as a substitute for them is selling you a gap. The argument is narrower and, we think, harder to dismiss: those controls were never designed to answer whether your accounts payable clerk will pay the changed bank details, and most organizations have no instrument that answers it either.
It is also not a maturity badge. The score the console produces is assembled from measured signals, and a category with no data behind it is excluded rather than filled in with an optimistic guess. If an organization has never run a simulation, the phishing dimension says so instead of inventing a number. That makes the score less flattering and considerably more useful.
Where to start
If you are reading this as an individual, the scanner is free and there is a walkthrough of how to use it and how to read the result.
If you are reading it for an organization, start by finding out where you actually stand rather than buying a program. Human risk management explains the discipline this sits inside, security awareness training that works covers why the annual module fails, and the small business cybersecurity guide puts the human layer next to everything else a smaller organization has to get right.
CyberWatch AI is built by two co-founders who are brothers and working security analysts, and it is self-funded with no outside investors and no advertising business. That shapes the product more than it sounds like it should. The detection logic comes from attacks seen landing on real people in operational work, and the only thing the product has to answer to is whether it is useful to the person using it.
Frequently asked questions
What is CyberWatch AI?
It is two things built on one idea. CyberWatch AI Scan is a free checker for a suspicious link, message or screenshot. The security console lets an organization test how its people respond to realistic phishing, train the gaps the tests expose, and track one score over time.
Is the scanner really free?
Yes. Checking links, messages and screenshots is free, with no daily limit and no account required. A free account keeps your scan history so you can look back at what you checked.
Does CyberWatch AI store what I scan?
The content itself is analyzed and discarded. Only the scan type, risk score and a short summary are kept. In an organization, administrators never see a check you did not choose to report.
Do phishing simulations collect passwords?
No. A simulation measures whether people click. Nothing in a simulation asks for a password, a card number or any credential, and the code refuses to send a draft that tries to.
Is this surveillance of employees?
No. The console is built to show security metrics, not private content. Leaderboards are anonymous by default, administrators cannot read checks an employee kept private, and the point is learning rather than blame.
Does it need an email-system integration?
No. Training and simulations work without a mail-server project. You invite people by email, group them into departments, and can run a first campaign the same afternoon.
Does it work outside English-speaking countries?
The scanner reads the content itself rather than matching a list of known scams, so it works in any language and for any country, including attacks only days old.
Who builds CyberWatch AI?
Two co-founders who are brothers and working security analysts. The company is self-funded with no outside investors, and there is no advertising business behind it.
Sources
- Phishing attacks: defending your organisation, UK National Cyber Security Centre
- Phishing, Technique T1566, MITRE ATT&CK
- Business Email Compromise, FBI Internet Crime Complaint Center
- SP 800-50 Rev. 1, Building a Cybersecurity and Privacy Learning Program, NIST Computer Security Resource Center
- Cyber Security Toolkit for Boards, UK National Cyber Security Centre
- Recognize and Report Phishing, US Cybersecurity and Infrastructure Security Agency
- Human-Centered Cybersecurity, NIST Computer Security Resource Center


