Can someone hijack my AI agent with hidden instructions?
An agent that reads email, web pages or files will also read instructions someone hid there. A planted file has already been enough to hijack coding agents.
We try the tricks attackers use on your agent and close the paths that work.
What you get
- A set of planted instructions tried against every place your agent reads from.
- A report of what it could be tricked into doing, worst first.
- Fixes for the worst paths, like tighter tool scopes or a confirmation step.
- The attack cases left behind as tests you can re-run after each change.
Price, timing and the guarantee
Prompt injection test
$499
Done within 4 business days of getting access to a test copy of your agent.
If we cannot run the tests and hand over the report and fixes listed, you get a full refund.
Rather talk first? Book a 15-minute call.
Payment runs through Stripe. Matt Turley will email you within one business day to set up a test copy of your agent we can attack safely. We never ask for a password by email.
Who does the work
Matt Turley does this work himself, with AI agents doing the legwork, and he checks everything before it reaches you. Continuum builds and looks after software for small businesses and founders. Questions before you buy: hello@uxcontinuum.com. Everything we offer.