Free
$0
One person, trying it out.
- 20 checks a month
- The same checks as the paid plans
- Every check on the record
- After 20 checks in a month, messages go through with a warning that they weren’t checked
AI protection for small business
SpadaShield checks what your team types into ChatGPT, Claude, Copilot and other AI tools, and stops client data and secrets before they reach the AI.
You set the rules. Every check goes on the record.
Free for one person, 20 checks a month. Works in Google Chrome. Setup is personal for now: we set up your workspace with you.
The problem
A tax ID in a ChatGPT prompt. A client’s card number in a Copilot chat. An API key pasted into Claude to fix a script. It happens in good faith and in seconds, and once it’s sent it sits on someone else’s servers.
SpadaShield puts a check in that moment, before the message goes. If it holds something that shouldn’t leave, it stops, and says what it found.
How it works
Install the SpadaShield extension and sign in with your Microsoft work account. We help with the install.
Use ChatGPT, Claude, Gemini, Copilot or Perplexity like always. Each message is checked before it’s sent.
If a message holds client data or a secret, it isn’t sent, and you see what was found. Flagged, never guessed. Fix it and send again.
What it catches
Common patterns of client data and secrets, plus the words only your firm knows.
You set the rules
Add client names, matter numbers and phrases that must never go into an AI tool. They’re checked like everything else, and a message they stop is kept without its text.
Each check is written to a tamper-evident record: every entry is chained to the one before it, so a changed or missing entry shows. Export it any time and check its signature offline with our standalone checker.
If our service can’t be reached, the message isn’t sent, and you’re told why. Nothing slips out unchecked because a connection dropped.
Ask, and we delete your workspace: from our live systems within 30 days and from every backup within 60.
What it doesn’t do
These are its limits today.
Pricing
Start free, or bring your team with a 30-day pilot. Full pricing.
$0
One person, trying it out.
$25 per user / month
$99 a month minimum, which covers up to 3 users.
$750 once
The full $750 comes off your first year if you stay.
From $2,000 per month
For larger firms, with terms set together.
Prices in US dollars.
Questions
Seventeen, in Google Chrome: ChatGPT, Claude (claude.ai and claude.com), the Anthropic Console, Gemini, Google AI Studio, NotebookLM, Google AI Mode, Microsoft Copilot (the consumer site), Perplexity, Poe, You.com, Character.AI, Meta AI, Mistral, DeepSeek, Kimi and HuggingChat.
It doesn’t cover AI apps installed on the computer, AI built into Office or email, or other browsers.
No. The checks run on our own servers. One exception: if a message contains a citation, the citation text alone is looked up in public research indexes to see whether it exists. Details on the security page.
It matches patterns, so some things get through: a client’s name with no title in front of it, an account number written in an unusual way, or a client described in plain words. It knows US formats, such as Social Security numbers, not other countries’ ID numbers.
The message isn’t sent, and you’re told why. It doesn’t go out unchecked.
For each check: what was found, what happened, who sent it and when, and the first 500 characters of the message, including drafts checked as you type. A message held by one of your firm’s own rules is kept without its text. More on the security page.
Not yet. There’s no SOC 2 report and no HIPAA business associate agreement today.
On Windows PCs joined to a company domain, IT can install it for everyone with a policy. Otherwise we help install it on each computer.
For the rest of that month, messages go through with a warning that they weren’t checked. Business has no monthly limit.
Or bring the whole team for 30 days with a person from NorthGate on email the whole time.