Security and data
Plain answers to what your security and risk teams will ask, including the parts that aren't built yet.
Running now works today. Pilot we're building with our founding clients. Planned comes after the pilot.
Your data
For Brand Protection: your brand names, your official web addresses and who should get alerts. Everything else Lookalike Watch reads is public: certificate logs, lists of newly registered web addresses and the suspect pages themselves.
Inbox Watch and Payment Watch will need access to your email or accounting system. We'll ask for it only when you add them, and only for what each one needs.
Running now Brand Protection Pilot Inbox Watch, Payment Watch
For a lookalike website: public facts about it, such as its address, its age, what its page says and a screenshot.
For an email: who sent it and whether it passed the sender checks (SPF, DKIM and DMARC), the sites its links point to, the subject, up to 5,000 characters of its text, the names and types of any attachments, and sometimes a screenshot of the front page a link points to. Attachments themselves, the full headers and the images in the email never reach an AI model.
Payment Watch is designed so that no AI model sees your bank transactions. Plain rules decide.
Running now Websites, Scam Check Pilot Inbox Watch, Payment Watch
Your account and case data, and the screenshots that are your evidence, are stored in Cloudflare's Oceania region. Our collection runs on our own server in Sydney. The AI models we use today run outside Australia, mainly in the United States, through Anthropic and OpenRouter. The privacy policy lists every provider and where it handles data.
Later, we'll run the AI models through AWS in Sydney and Melbourne, so what they read stays in Australia and the company that makes the models can't see it.
Running now Storage in Oceania Planned AI in Australia
We keep the evidence of each fake, such as screenshots, page copies and dates, because it's your record. Inbox Watch keeps a record of each email it checks: the sender, the check results, the links, attachment names, the verdict and the AI's reasons, plus the subject for the small sample our team reviews. It never stores the email itself or its attachments. Our privacy policy sets twelve months for messages people forward to Scam Check.
We haven't set deletion schedules yet. They're planned with our lawyer: shorter for anything from email, longer for evidence, and everything about your organisation deleted when you leave.
Planned Retention periods and deletion on exit
Decisions and control
It sorts every case and closes the obvious ones that aren't fakes. One in five of those it clears while the page is live goes back to a person to check. A person decides before any takedown request or alert goes out, and every decision is recorded with who made it and when.
Next, the AI will alert you and update your blocking list on its own for clear-cut fakes. It runs in shadow today, measured against every person's decision, and switches on only when its record earns it. Takedown requests and reports will always wait for a person.
Running now A person decides every action Pilot AI acting alone on clear-cut fakes
You'll choose, for each module, how much a model may see: nothing, so plain rules decide; only the facts the rules produce, such as addresses, dates and check results; or the content as well, processed in Australia. We'll show you what each choice costs in accuracy before you pick.
Planned
Your people sign in with their own email and see only your organisation's cases, never another client's. Someone you remove loses access at once. Each person's role decides whether they can only look or can also decide.
Your blocking list is the one thing reached without signing in. Your tools fetch it through a private link, which you can replace at any time, and every fetch is counted.
Running now
Takedown requests to the company hosting a fake and the one its address was bought from, and reports to Google and Microsoft so their browsers warn people. Each carries the fake's address and evidence and names the brand it copies. They go out in Sixth's name, after a person approves them.
Later: requests in your own name under a letter of authority, once a lawyer has checked them, and every confirmed fake shared through the anti-phishing industry's exchange so email providers and security companies block it too.
Running now In Sixth's name Planned In your name, industry sharing
Assurance
Not yet. ISO 27001 certification and an assessment against the Australian Government's Essential Eight are planned. Until then, we'll answer your security questionnaire in full and walk your team through how Sixth is built.
Planned
Passwords and keys are kept in a managed password vault, or as secrets on Cloudflare, and reach each job only when it runs. Every part of the engine is health-checked, and our team gets an email when one has a problem.
Running now
The console and your blocking list run on Cloudflare's network. Watching for new fakes runs on our own server in Sydney today.
As our first clients come on, we'll move it to an Australian cloud region with a standby, and write incident notice and exit terms into your contract.
Planned Cloud collection, contract terms
Cloudflare hosts the site and the console, stores your data and opens suspect pages in a sealed-off browser. Anthropic and OpenRouter run the AI models. The privacy policy lists every provider, what it does and where.
Running now
We'll answer it in full, including the parts that are still being built.