Fortress
GitHub
A new stealth Chromium fork called Fortress fixes browser fingerprints at the engine level, not with JS patches. It slips past CreepJS, Cloudflare, and Akamai bot checks that catch normal scraper tools.
Based on reporting by GitHub — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Most anti-bot workarounds are duct tape. They patch navigator.webdriver in JavaScript, fake a WebGL vendor string, and hope nobody looks too closely. Fortress, a new open-source Chromium fork from a team calling itself Tilion, skips the duct tape entirely and rewrites the browser's C++ getters directly, so the fingerprint a site reads is native code all the way down.
That distinction matters more than it sounds. Detectors like CreepJS don't just check what navigator.vendor returns, they check whether the function returning it is actually native. Call .toString() on a JavaScript shim and it exposes its own source. Grab the same primitive from a fresh iframe or a Web Worker and reuse it against your patched object, and the trick collapses instantly. Fortress avoids all of that because there's no shim: the getter itself is compiled into Blink, so it reports [native code] everywhere, in every realm, because it genuinely is native code.
The project ships as a drop-in binary that exposes a normal Chrome DevTools Protocol endpoint on port 9222, so existing Playwright or Puppeteer scripts connect without touching a line of automation logic. Tilion's demos, run against live systems rather than sandboxes, show the engine clearing a real Cloudflare Turnstile challenge, going all-green on bot.sannysoft.com, and getting past Akamai's Bot Manager on sites like American Airlines and Lowe's using the same residential IP that a stock headless Chrome gets blocked on. The claim, repeated a few times in the docs, is that if you're still getting blocked after switching to Fortress, the problem is your proxy, not your browser.
The release also comes with an MCP server, so agent frameworks like Claude Code or Cursor can call tools such as fetch_protected_page or crawl_site directly instead of wiring up CDP by hand. Twenty-nine tools in total, aimed squarely at the current wave of AI agents that need to actually reach the pages they're told to summarize or transact on, not just get bounced by a 403.
Worth noting: Fortress is BSD-3 licensed, ships 34 auditable patches against upstream Chromium, and rebases monthly, so anyone worried about a black-box binary spoofing their traffic can read the diffs and rebuild it themselves.
My take — AI-written commentary, not fact-checked reporting
This is the logical endpoint of the scraper-versus-detector arms race: once your adversary checks whether code is native, the only durable answer is to actually make it native, and I'd bet most JS-based stealth plugins are functionally dead within a year or two. I'm generally uneasy about tools built purely to defeat bot detection at scale, but an open, auditable patch set is a far more honest way to do it than another closed vendor selling you a black box and your trust.
Read more about this at: GitHub
Related stories
Cloudflare Introduces Kitesurf: An Agent-First Web Browser That Runs Entirely in V8 Isolates on Cloudflare Workers
MarkTechPost · 1 month ago ·
9
Firefox in WebAssembly
Simon Willison's Weblog · 2 months ago ·
16
Introducing Kitesurf: The agent-first browser that runs in V8 isolates on Cloudflare Workers
Cloudflare Blog · 1 month ago ·
43