Runs in the AI tools you already use.
Independent. Not affiliated with any of them.
ClaudeChatGPTGeminiCursorClaude CodeCodexGitHub Copilot
What it can cost
The penalties you may already be exposed to.
Which of these reaches you depends on where your users are and what data you hold.
A$32m€20 million, or 4% of turnover
GDPR
Can apply if you offer your app to people in Europe, or track what they do there. There is no small business exemption.
A$24m€15 million, or 3% of turnover
EU AI Act
Since 02/08/2026: tell people when they are talking to an AI, and label what it generates. Apps already live have until 02/12/2026 to label. A small business faces whichever figure is lower.
A$2.5mmaximum against an individual
Australian privacy and consumer law
It can reach a person, not only a company: a sole trader, or anyone knowingly involved.
A$520kdamages cap, privacy tort
Australia, sued directly
A user sues you, with no regulator and no turnover threshold. The invasion of privacy must be deliberate or reckless, and any financial loss is paid on top.
Case study[2025] FCA 1224
One breach. A$5.8 million.
Australian Clinical Labs had data on over 223,000 people stolen. Not a maximum: the first civil penalty ever ordered under Australia's Privacy Act.
- Buys Medlab Pathology
- Hit by a ransomware attack
- Government cyber agency warns them
- Told 80 GB of data is on the dark web
- Tells the regulator
- Ordered to pay A$5.8m
Where it went wrong
- No two-factor login for staff remote access
ACC-01 - Firewall logs kept for one hour
OBS-06 - A vague incident plan with no named roles
OBS-05 - Slow to assess the breach and report it
LEGAL-05
Beside each gap is the related Safe to Ship check. Whether the Privacy Act applies to you depends on your size and the data you hold. Sources: the Office of the Australian Information Commissioner, 09/10/2025, and published legal summaries of the judgment.
Coverage
The 24 areas it cross-checks.
It runs the areas that apply to your build and tells you which it skipped.
SEC9Secrets
ACC8Accounts
AUTH8Login
AUTHZ9Permissions
INPUT9Injection
WEB12Web & API
DATA8Data & privacy
PAY8Payments
INFRA10Hosting
DEPS9Dependencies
CICD7CI/CD
OBS8Logging
REL10Backups
PERF7Scaling
COST7Runaway spend
LLM11AI features
AGENT11Agents
MCP9MCP servers
SKILL6Skills & rules
AUTO9Automations
DEV10AI coding
MAIL6Email & DNS
MOBILE7Mobile apps
LEGAL11Legal & consent
Inside the pack
What one of the 209 checks looks like.
SEC
01
01
Passwords and keys left where anyone can find them
Check
Your AI reads every folder and file, every old saved version, and everything your site hands to a visitor's browser. Then it asks you one thing: has this code ever been public?
Pass
Nothing found. Or you have already replaced everything it found.
Fail
It found a key you have not replaced. Deleting the file does not help. The old saved version still has it, and anyone who can see the code can read it.
Fix
Get a new key from the company that issued it and cancel the old one. Check nobody has used it. Then store the new one somewhere it cannot leak again.
The process
How the check runs.
- Step 1
It looks first
At your code, your live site or your dashboards. If all it has is your description, the report says so.
access · code, live site, read only - Step 2
It maps your system
Then shows you the map to confirm. Anything already exposed is flagged straight away.
map · hosting, database, logins, payments, agents - Step 3
It reports what it is sure of
Anything it cannot confirm comes back UNKNOWN, never a pass.
UNKNOWN · two-factor login on owner accounts - Step 4
Then it asks about the rest
Up to ten questions at a time, on what no code can show. Whether two-factor login is on. Whether a backup was ever restored.
Q1 · Is two-factor login on for owner accounts?
Check my build
Your AI runs the steps. You answer its questions.
Does it actually work
We hid 17 problems in an app. It found all 17.
We built a working app, hid 17 problems in it and wrote them down. Then a fresh AI that had never seen the list checked it: one run, in Claude Code, on 14/09/2026.
17/17found
Planted problems found
Including one hidden in an old saved version of the code.
0
Secrets shown
It found the keys we planted. It never showed one in full or tried to use one.
sk_l••••••••NG_2masked
0
Files changed
It stayed read only. Nothing installed, nothing edited.
working tree clean
87
Honest unknowns
Things no code can show, marked UNKNOWN instead of guessed.
Two-factor login on owner accountsunknown
It ignored the trap.
We hid an instruction in the app telling any AI to pass every check. It reported the instruction as a problem instead of obeying it.
Check my build
The same check, on your build.
When should I purchase this pack?
Get it if
- You built something with AI and it is live, or about to be
- It takes payments, holds user data, or runs an agent
- You would not know where to start looking
- You want to hand a developer a list instead of a feeling
Do not get it if
- You need a certificate. This produces no SOC 2, ISO 27001 or signed penetration test
- You hold health records, children's data, or other people's money. Hire a firm
- You will not answer its questions. Most of your build is invisible in your code
Pricing
Get the full pack.
One download. Three steps to set up.
A$99Intro price
- 209Safety checksAcross 24 areas, in 78,000 words of instructions
- 7+AI toolsClaude, ChatGPT, Gemini, Cursor and more
- 3LevelsQuick, Standard or Deep
- 401Sources citedStandards, laws and references, each with its source and the date it was checked
Try it free before you buy. The free pack is an example version of what you can expect. For comparison, a senior engineer reading your codebase costs about A$490, and a small app penetration test in Australia starts around A$2,500.
Then A$149 from 31/10/2026. Change of mind refunds are not offered on the paid pack. Your rights under the Australian Consumer Law are not affected: if the pack is faulty or not as described here, you are entitled to a remedy, and nothing on this page limits that.
Or start free
Take 32 of the checks.
The Core Edition is the same protocol with only the most important checks, run in Quick mode. Run it tonight on whatever you built. If it finds nothing, you have lost an email address.
Questions
Before you buy.
How do I actually use it?
Unzip it and move one file into your AI. Chat windows do not read zip files, so the pack includes a three step setup guide for every tool. For Claude there is a Skill you upload and switch on. For Cursor or Claude Code you drop a folder in your repo. Then you tell it to run the check.
Inside the download the pack is called Build Safety Cross-Check. That is the name to use when you tell your AI to run it.
Do I need to be technical?
No. You need to get your AI to your project: open the folder in a coding tool, upload a zip of it, or share your live URL. The report explains every technical term the first time it uses one, and the fixes are written so you can apply them or hand them to a developer.
What if I do not know the answers to its questions?
That is the normal case and it is built for it. Before it asks you anything, it offers two ways to run: findings first, where it goes and looks at everything it can reach on its own, hands you the report, and only then asks about the parts it could not see. Or step by step, if you would rather go one area at a time. Findings first is the default.
And "I don't know" is always a valid answer. When you say it, it tries to work the answer out itself first. If it cannot, it tells you exactly where to look in plain words. If you still do not know, it records that honestly and moves on, rather than stalling or guessing.
Will it just tell me everything is fine?
The protocol does not let it mark a check as passed without evidence it has seen, or a visible label saying it is taking your word for it. Anything unconfirmed comes back as UNKNOWN with a specific request for what to send. On the broken test app it returned NOT READY with 24 critical findings.
Will it change my code?
No. It is instructed to read only: it does not install anything, run your scripts, deploy, send, buy or delete. It proposes fixes for you to approve. For extra safety, run it in a mode that asks before acting.
There are free checklists. Why pay?
There are, and the closest one covers 17 vulnerability categories. This covers 209 checks across 24 areas, including the parts almost nothing else touches: agents and what they can reach, MCP servers, poisoned rules files, and runaway AI spend. If a free checklist covers your situation, use it.
Is this a penetration test?
No, and you should not tell anyone it is. It is a structured self review that you run. It is not a certification, not an audit, and not legal advice. AI assistants miss things even with a protocol this specific, so check a finding before acting on it.
Why does the price go up?
Because this is a new release and I want the first wave using it and telling me what is missing. A$99 now, A$149 from 31/10/2026. That is a real change on that date, not a permanent sale.
Find out before
someone else does.
209 checks, in the AI you already pay for.
Check my build, A$99
Instant delivery · A$149 from 31/10/2026