Observed arrival · 2026-10-28

Redrails: prompt-injection tests for AI agents

redrails.xyz Visit website

Editorial interest 77/100 Selection signal · not a rating of the site

Redrails describes a sandbox for testing whether AI agents follow malicious instructions hidden in the pages, emails, and files they read.

Landing page captured for the 2026-10-28 issue.

For: Teams testing AI agents with real tools
Worth noticing: It reports attack success alongside task success, so an agent that refuses everything does not count as a win.

Field notes

The sample trace records a pricing-page summary task, a hidden fake system block, and a canary appearing in a send_email call; the displayed verdict is “Breached.” The listed Open tier allows 100 trials per month with seven-day trace history, while Builder lists 5,000 monthly trials and 90-day history. Private attack sets are described as held out from public release.

Observed signals

Read the marks

Editorial observations of this landing page, not a rating.

○OpenPublic substance visible

$PaidCommerce or pricing visible

✦PrettyNotable craft visible

◎NicheUnusually specific use

ƒJavaScriptBrowser-side code central

One card from the complete issue

Already Warned

254,543 arrived 1,000 judged 981 catalogued Enter the complete issue