Prompt-Injection Red-Team for Tool-Using Agents
For businesses whose AI agent reads emails, pages or tickets and can take actions. We attack it like a real intruder would, show every trick that worked, fix each hole and keep testing on every release.
- Free scoping assessment first
- Inspected before it goes live
- Monthly or every 3 months, or yearly
- 1–3 days build time, once your material is in
What you get
- Set of attack tricks made for what your AI assistant can do and read
- Proof of each trick that worked and what damage it could do
- Fixes: keep outside content separate, limit what the assistant may do, check its answers
- AI filter that checks incoming content for hidden instructions
- Attack tests run automatically before each update
- Follow-up report showing each trick is now blocked
How your order runs, after you choose a plan
Questionnaire
You answer a short questionnaire about your business, your goals and what this service needs.
Five sample concepts
We prepare five sample dashboards from your answers, each showing your numbers in a different way.
You choose or mix
Pick one, or tell us what to take from two of them.
Upload your material
Send what this service works from — files, data, access or material, depending on the service. We use only what you give us.
We build it
Our specialist AI team builds the complete service from your choice and your material.
Quality inspector
An automated inspector checks the work. Anything that fails is rebuilt before you see it.
Live
Your dashboard opens from your account and updates on a schedule.
The five samples usually arrive within minutes. Each service's build time is counted from the day your questionnaire and material are complete.
How it works
What it does for you, every day, once it is live.
- 1Runs the attack suite against every new release before deploy
- 2Generates fresh hidden-instruction attacks based on your agent's tools
- 3Screens incoming content for planted instructions before the agent reads it
- 4Checks agent outputs and blocks risky actions for human approval
Connects to
Connected to the tools you already use.
How we deliver it
Choose your plan
Every plan includes the complete service. A bigger plan includes more usage.
Pilot
Start small- The complete service is included in this plan.
- Less than $3 a day
- Included every month — Attack attempts tested: 50More when you need it: $18 for 20 extra ($0.90 each)
- 2 care rounds a month — changes, fixes and help
Start
- The complete service is included in this plan.
- Less than $5 a day
- Included every month — Attack attempts tested: 100More when you need it: $18 for 20 extra ($0.90 each)
- 2 care rounds a month — changes, fixes and help
Growth
Recommended- The complete service is included in this plan.
- Less than $9 a day
- Included every month — Attack attempts tested: 250More when you need it: $18 for 20 extra ($0.90 each)
- 2 care rounds a month — changes, fixes and help
Scale
- The complete service is included in this plan.
- Included every month — Attack attempts tested: 600More when you need it: $18 for 20 extra ($0.90 each)
- 2 care rounds a month — changes, fixes and help
Bigger than Scale? Enterprise is quoted after a free assessment.
- Built on your business — your data, rules and tools — by our specialist AI team, not a generic template.
- Every delivery passes our quality checks before you see it: tested, working and documented.
- On every plan: we keep it live and running, improve it each month from its own results, and your monthly care rounds cover changes, fixes and help.
Included in the price
- The service built for your business and running (or delivered)
- A live dashboard on your own data
- Your files, sheet or system connected
- A plain-words summary each time it updates
- Changes after delivery: rounds of care included
Extras
Tick what you want on top; the total updates as you go.
What we need from you
We use only what you give us — we never invent your details.
- What can your agent read, and from where?
- Which tools and actions can it trigger?
- Can we test against a staging copy?
- Which actions must always need a human?
- Where do your releases run?
Your data stays yours.
The promise, in plain words — and each part is built into how we work, not just written here.
Encrypted, per customer
Everything you send us is encrypted with keys that belong to your account alone.
Used only for your service
Your files and answers are used to build your service — never for another customer's work.
No outside AI without approval
Data you mark confidential is not sent to any outside AI service we have not approved. By default it is handled only by our own tools.
Delete everything, any time
From your account you can delete everything you sent us and everything built from it. Your keys are destroyed, so even backups become unreadable, and you get a deletion receipt. Only the invoice records that accounting law requires are kept.
Add-ons & upgrades
Start with the build, then add what you need — anytime.
Built for your sector
Ready to see it built on your business?
Start with a free scoping assessment — we scope it and quote before you commit.