A complete walkthrough of the working product, covering every module in the order an incident moves through them. The organisation shown is fictitious, so the recording can be shared without restriction.
Every identity, domain, address and network detail in this recording is fictitious. No customer environment appears at any point.
A filter owns the moment of delivery. The weeks before it, and the months after, are where breaches actually happen, and where a small team is usually on its own. Step through all five.
A domain is registered to look like yours: a swapped letter, a lookalike character, a different ending. Mail gets switched on. Somebody tests whether they can send as you. All of it visible, all of it before a single message lands.
Finds the copycat domains someone has actually registered, and which of them are set up to receive mail.
Turns the daily reports providers send back into a readable list of who sends mail as you.
Watches leak sites for your name and your suppliers', so you hear it in hours.
Mail, a Teams chat with an outside company, or a file in a shared library. The platform never waits to be told.
Checked within seconds of arrival. Watch only, or quarantine on its own.
Reported mail reviewed and detonated safely, laid out for a decision in one screen.
Microsoft, Check Point and Kaspersky held mail, in one list.
One click in Outlook or on iPhone. Training sims get credited, not queued.
Modern credential theft relays the real Microsoft page through the attacker's server, so the approval on your phone gets used by them. Filters never see it.
A hidden marker in your sign-in branding. A relayed page loads it and gives itself away.
Learns each person's normal, flags what doesn't fit, reads next to the tripwire.
One person reported it, forty got it. Find every copy and remove it.
Mail gets copied out. A rule hides the bank's replies. An app approved once keeps reading mail after the password change. This is where invoice fraud actually lives.
The clean-up list: forwarding, hidden rules, approved apps, extra access. Fix button on each.
Reads the account's own sent mail. A phished mailbox sends the next wave.
A password change is useless if they still hold a live session. Spots the reuse.
Watches your sign-in rules for drift, and who changed them.
Read-only sweep of the office directory, including which staff reuse a password that has already leaked.
The standing access applications hold as themselves, with no user and no sign-in to notice.
Certificates survive a password reset. Lists them per person, cancels on approval.
Every stage above is something that already failed. The only honest way to measure the last one is to attack it yourself.
Eight real techniques, 154 ready-made exercises, taught at the moment of the mistake.
A Teams or Zoom join screen, or a live cloned voice on the call.
Email signatures. One approved layout on every mailbox, filled in from your directory.
A Teams link from a partner never passes your email security at all. A shared file gets opened by everyone.
Chats with outside organisations, checked as messages arrive. Clean chat is discarded, never stored.
It never silently alters Teams. Cutting off a compromised partner is a separate, recorded, reversible action.
Shared libraries and Teams channel drives, watched continuously rather than on report.
A customer's document is never uploaded to a third-party scanner to decide whether it's safe.
One bad file or message becomes a sweep across every library and conversation.
Microsoft dropped the Teams export metering in August 2025. No special arrangement needed.
What matters on a bad afternoon is what you can do without waiting for a rule to agree, and how fast one click can undo it.
By sender, subject, phrase, link or attachment. Watch hits arrive, stop early.
Open any result safely, or take it away as a file. Nothing goes because a tool decided.
Bulk delete from every mailbox. Soft while unsure, purge when certain.
Restore from quarantine, deleted items, or recoverable items most tools ignore.
Release, delete, and answer whoever has been chasing you about it.
Throwaway browser in a container. Can't copy anything back out. Self-destructs.
Staff report a suspicious message from the toolbar in one click. Real threats land in your queue; a training simulation is scored and the person told straight away.
Import a saved message from another company. Same analysis as everything else.
Searching and purging are different permissions. Every action is attributed.
Eight containment actions sit as buttons beside the finding that justifies them. No scripting, no portal hunting.
Push alerts to the browser the moment something lands, so nobody has to sit watching a console.
Phishing is only the delivery. What the attacker is shopping for is a credential that still works. A great many of them leaked somewhere else, years ago, and were never changed.
See which of your staff are signing in today with a password that is already published, before somebody else gets round to trying it.
The breach lists come to you. Your directory never goes out to somebody else's checking service, and no password is ever read, sent or stored.
A read-only sweep of the directory for the openings an attacker uses once they are already inside and looking for a way up.
Change the password and the attacker may still be holding a live token that works. Spot the stolen session being replayed, and cut it.
A password can be typed into a convincing fake. A key sealed into the machine itself cannot be copied out, forwarded, or handed over by a helpful employee. That is what makes certificate sign-in the end of the problem rather than another control.
A single misconfigured certificate template lets anyone allowed to enrol request a certificate in somebody else's name, an administrator's included. It is quiet, it is permanent, and almost nobody checks for it.
A certificate issued to a person keeps working long after their password changes. Most teams have never once looked at the list.
Your sign-in policies, scored against a zero-trust baseline, with every change attributed to whoever made it.
Not a model on its own. A cascade of purpose-trained models wired into custom inference loops, with detonation, optical reading and sender-authentication signals fused into a single verdict. Engineered to run on an ordinary server CPU: no GPU, no accelerator, no cloud, no API key.
A small, fast transformer trained purely on phishing language sees everything first. Most mail is settled here and nobody is interrupted.
Uncertainty escalates itself to a far larger model before anyone is told they have been attacked. The cascade decides when that is worth the compute.
A text message and a long email do not read alike. Add a trained specialist per shape and let the ensemble weigh them.
Security AI usually means renting somebody else's accelerators and posting your mail to them. This cascade was engineered to fit a plain virtual machine. Quantised weights, batched CPU inference and a routing loop that spends the big model only where it changes the answer.The appliance runs on ordinary vCPU and RAM, with no CUDA, no accelerator passthrough and no per-token bill. Adding a specialist model is a download, not a hardware project.
A lure first sent this morning matches no blocklist and no signature, so most filters wave it through. Nothing here depends on having seen the attack before. The pipeline reasons about the message itself: how the language is built, whether the sender really holds the domain, where the link lands once followed, what the attachment does when opened, and what the picture says.Signature and reputation checks are contributors to a score, never the deciding vote. A verdict can be reached with every reputation source switched off.
A brand new address has no reputation to check, so the page is fetched and detonated to see what it actually asks for.
Impersonation is judged against your real domains, your branding and your sending sources, not a global list of known-bad names.
If a flawless message does get through, the relayed sign-in page and the odd token redemption still give the attack away.
If any stage cannot reach a confident answer, the message is held rather than delivered. Uncertainty is not treated as safety.
It may shortcut a verdict of "safe". It may never shortcut "phishing". A false accusation costs an analyst's afternoon and teaches your staff to ignore the tool.The shortcut and escalation confidence levels are both settings, and the never-shortcut-an-accusation rule is auditable.
An email is not just text and links. It reads the screenshots, the logos, the attachments and the QR codes too, so nothing hides in a picture.
Which words drove the verdict. "The model said so" isn't defensible to a manager.
Every verdict your analysts correct becomes training data. It gets better at your company, not at a benchmark.
The engine works with the network unplugged. No AI provider, no subscription, no telemetry. The trained model is yours.
Watched from beneath the operating system, where evasion gives up.
Everything a file touches, on your hardware.
Always on, the default for every attachment.
Already run an on-prem detonation box? It plugs in.
Open a hostile link by hand, safely.
Built for one message, destroyed after.
Shares: the address and your IP.
Shares: the file itself. The most invasive.
Shares: a hash fragment, not the address.
Shares: the IP only.
Shares: the indicator looked up.
You see the service, the data and the retention before anything is sent.
154 exercises across email, text, phone, QR and fake meetings. Teach in the one moment anyone is paying attention.
Caught by a QR code? You get the QR lesson. One page, one minute, not a module.
Clicking is noise. Speaking up is the behaviour you want, so that's what's ranked.
One campaign per department, with pretexts and senders that fit. Spread over hours.
A short sequence when they appear in the directory, with a follow-up later.
Select any technique for what it rehearses and how it is delivered.
A DMARC reader, a leak-site monitor and an account scanner. Here's the honest case.
Look at what the reports contain: every service on earth that sent mail claiming to be you, and whether it was believed. That's a list of who impersonates you.
It also catches what you'd never otherwise hear about. Phishing aimed at your customers using your name never touches your mailboxes, so no filter will ever show it to you.
Two things, and deliberately nothing more. Convincing scams are built from leaked data. The invoice quoting your real purchase order came from somewhere.
And most of those break-ins started with phishing. Appearing on a leak site is your own anti-phishing work being marked from outside.
Watcher answers the question every phishing incident ends on: the password went somewhere, so what did they do with it?
Deleting the email isn't the fix. The fix is the forwarding rule, the rule hiding the bank's replies, the app still reading mail after the reset. That list is invoice fraud, and a tool that only reads email cannot see it.
Nothing here asks you to tear out what you have already bought. PHISHZ reads its verdict, works its quarantine over the API, and takes on the stages no mail filter was ever built for, whether yours is a gateway in front of Microsoft 365, an API product sitting beside it, or Microsoft's own.
Every message records which product handled it on the way in and what that product decided, so you can see exactly where something got through.
Work Microsoft, Check Point and Kaspersky quarantine from the same screen. Release, delete and answer whoever is chasing you, without three portals and three logins.
Your email security and our pipeline look for different things. When the verdicts differ you see both, side by side, and you decide, rather than finding out from the person who got phished.
Bring the threat-intelligence sources you already trust. They stay switched off until you ask for them, and the engine never needs them to reach a verdict.
What the platform learns does not stop at the console. Hand it to the firewall and the endpoints you already manage.
Nothing in front of Microsoft 365? Microsoft 365 and PHISHZ is a complete posture by itself, and it is what a lot of small teams actually run.
Sixteen on this page, and more inside them: Watcher alone holds Password Guard, CA Guard, TokenGuard and app permissions. Filter by the problem you are solving.
One list of what needs you today, from every module. No checking five screens.
Review and detonate a reported email safely. Real sender, real link destinations, a safe picture of the page.
Search every mailbox at once, remove what you find, restore if you overreached.
Search · Preview · Bulk delete · Restore · Notify
Checks mail within seconds of arrival. Also Teams chats and shared files.
Flagged mail · Teams · Files · Mailboxes · Action history
The clean-up list after a compromise, with a fix button on every item.
Monitored accounts · OAuth apps · Password Guard · CA Guard · TokenGuard · App permissions
Eight techniques, 154 exercises, taught at the moment of the mistake.
Campaigns · Email templates · SMS · Voice & IVR · Landing pages · Organisations · Leaderboard
Email signatures, standardised across every mailbox, with an alert when someone edits theirs.
Users · Classic devices · Monitoring · Group assignments · Templates · Photos · Add-in
A readable list of everyone sending email as your company.
Copycat domains, and which are set up to receive mail rather than just host a page.
Catches the attack that defeats approval prompts: your login page, relayed.
Leak sites watched for your name and your suppliers'.
Leak sites · Onion index · Breach feeds · Watchlist
Runs certificate sign-in as a phishing-proof replacement for passwords: who holds one, whose key is hardware-bound, and which templates are unsafe.
Certificates · TPM attestation · Template abuse (ESC1) · Autoenrolment · Cleanup · Revocations · CRL
Microsoft, Check Point and Kaspersky held mail, reviewed in one place.
Microsoft · Check Point · Kaspersky · Release requests
A throwaway browser plus the file and link checks the whole platform uses.
Browser sandbox · URL scan · File analysis
How retention, journaling and activity logging are really configured.
Every action attributed, roles that limit who can delete, honest health checks.
Services · Host OS · Audit log · Backups
Same team, same principles, same appliance model. The names are not subtle: PHISHZ is phish zenith, ENFORZ is enforce. Each is free, each runs on your own hardware, and neither needs the other.
Mail, Teams, shared files and the channels no filter reaches. Caught before delivery where it can be, remediated after the click where it cannot, and your people rehearsed against the rest.
Eventually somebody clicks anyway. ENFORZ is Windows application control made survivable: whatever was downloaded does not execute, because it was never approved, and nothing had to recognise it first.
The lure is caught before it lands, the people are trained for the ones that get through, and the payload cannot start even if everything else fails. Two appliances, both yours, nothing shared with us.
That makes it a target, and it's built on that assumption.
Internet-facing parts live on separate machines. The appliance only ever calls out.
Every credential, token and certificate lives in a sealed HashiCorp Vault. Never in a config file, never in an environment variable, never on disk in the clear.
A locked-down Linux base where every service runs without privileges, is walled off from the others, and cannot gain new rights once it is running.
Roles gate who can delete or release. Every action is recorded against a name.
The whole platform ships as one virtual machine for Hyper-V or VMware. The operating system underneath is hardened and patched as part of the release, so there is no Linux build for anyone on your side to get right, and no stack of dependencies to assemble.
The organisations with the least to spend on security are targeted with the same lures as everyone else. PHISHZ exists to close that gap. It is free, the source is open, and there is no commercial edition holding the useful parts back.
It is free because we believe in a more secure future, and you cannot build one by selling it only to the organisations that can afford it.
An organisation without a security budget is not a smaller target. It is an easier one, and attackers have worked that out. So there is no paid tier waiting for you later, no feature held back, and no point at which this starts costing money.
The installation is yours, the hardware is yours, and everything inside it is yours. Nothing is shared with us or transmitted to us: not a message, not a name, not a count of how often you use it.
No licence, no subscription, no per-mailbox charge, and no edition where the useful parts cost extra. All sixteen modules are in the one release. Your cost is the server it runs on.
Every detection rule and every action is in code you can read, change and audit before you trust it.
No telemetry, no usage reporting, no upsell path. Nobody profits from what your mail contains.
Ask who we are, how the organisation is funded, and why we give this away. We will tell you plainly. Just as happily: a walk through a deployment, the Microsoft 365 permissions it needs, or a straight answer on whether it suits your organisation at all. There is no sales process, because there is nothing to sell.
Spot the groundwork, catch what lands in mail, Teams and files, remove every copy yourself, close what is left behind, and train the people it was aimed at. One platform doing the work of a stack, for the kind of team that does not have a stack, or a night shift.