Observed arrival · 2026-11-10
AgentSafe checks URLs before AI agents read them
AgentSafe presents an API for scanning web pages for hidden prompt injection and phishing signals, with a verdict and cleaned Markdown as output.
- For
- Developers building agents that read untrusted web pages
- Worth noticing
- The sample response includes a threat type, severity, channel, evidence snippet, and clean_markdown field; JavaScript execution is not supported.
Field notes
Requests can be sent to /v1/scan, which the page says returns a verdict, risk score, threat records, and cleaned Markdown; /v1/check checks a URL or domain without fetching it. The examples identify hidden CSS, comments, alt and ARIA text, metadata, form destinations, punycode, and zero-width Unicode as inspection points. The page offers 200 free scans daily without signup and says unreadable pages are not charged; it also notes JavaScript-rendered content is not scanned.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue