When you supply `inject_css` or `inject_js`, that code runs in the same browser context as the page being rendered (not in an isolated JavaScript sandbox). You should treat injected snippets as trusted code.
To reduce cross-site and SSRF-style abuse from scripts that trigger new network requests, our rendering workers install request interception for captures that use a URL target or non-empty inject parameters. HTTP(S) requests from the page are checked against the same **public URL** policy used for the main capture (private IPs, metadata hosts, and similar targets are blocked where the check applies).
This does **not** provide a full VM-style sandbox: malicious or inefficient scripts can still affect CPU, memory, or the DOM inside the ephemeral browser session for that job. We additionally enforce plan tiers, request-size limits, and abuse controls (including optional restrictions on inject features for high-risk accounts).