Who builds SafePrompt, and how do you check the numbers?
SafePrompt blocks the messages that try to hijack your AI, before your model reads them. One call, and the attacks it finds never reach it.
About page last reviewed: 12 September 2026
Why SafePrompt exists
SafePrompt guards the kind of app that ships in a week. A contact form whose submissions an AI summarises for your inbox. A support bot sitting on a pricing page. An agent that reads your email and books the meeting.
In December 2023 a customer typed a few lines into a Chevrolet dealership's chatbot and got it to agree to sell a Tahoe for one dollar. In March 2026 developers talked Chipotle's support bot into writing Python, wrapped its back end in a tool anyone could install, and ran their coding work on Chipotle's compute, as Webb Wright reported at Gizmodo. Neither attack needed a password or a software exploit. Both were sentences typed into a text box.
That text box is what kept turning up in agency work. The message field on a lead form went to the model exactly as the visitor wrote it, and nothing in between read it first. SafePrompt is that check.
How can you verify SafePrompt's published numbers?
Every suite figure on this site renders at build time from the latest run of our public benchmark suite, which you can clone and re-run yourself, and every external benchmark row renders from the committed run record behind the dated benchmarks page. The comparison page quotes the dated runs it names. Current aggregate measurements are published on SafePrompt; dated external per-case results are linked from the benchmarks page. Exact-current per-case publication remains pending.
The SDKs are MIT licensed and the source is on GitHub. Read it and you can see exactly what your app sends us before you send anything.
What SafePrompt covers
SafePrompt screens submitted text for instructions that try to redirect the AI reading it. Your application enforces each verdict and authorizes actions for its users.
A stack trace, quoted ad copy or frustrated question may be ordinary application traffic. Injection detection is separate from content policy, and ordinary messages can still be flagged. Test your workload against the detection limits.
Who builds SafePrompt?
SafePrompt is built by Ian Ho, formerly a technical architect at eBay, and shipped by Reboot, Inc. in Las Vegas, Nevada.
Start on the free plan
10,000 free validations a month, on the same detection engine every paid plan runs.
Start with your email. No card.
Questions first? Contact us.
SafePrompt is a product of Reboot, Inc., Las Vegas, Nevada.