Vulnerability Safari: How a WebAssembly Flaw Bypasses the Same-Origin Policy

Anthropic Reveals Mythos and Project Glasswing: A Vulnerability Violating the Same-Origin Policy

When Anthropic presented Mythos and Project Glasswing, the reaction was immediate and polarized. Some dismissed it all as fear-based marketing, while others considered it a credible change in the threat landscape. Probably the truth lies in between. I wanted to verify it myself and, having recently gained access to OpenAI's Trusted Access for Cyber program, I decided to test it.

GPT-5.4 identified the bugs and helped assemble a working exploit chain, but it was not a simple prompt like "build me an exploit". Guiding the model required domain knowledge, iterative investigations, and the ability to recognize which paths were actually exploitable.

In modern browsers like Safari, exploitation is less about bug hunting and more about discovering bugs that still have a significant impact after multiple layers of defense.

The Bug That Violates the Same-Origin Policy

The bug I will talk about today falls into a more interesting category. The bug itself seemed contained and, in many ways, it was. It did not provide a path for remote code execution or sandbox escape. What it did, instead, was cross a different boundary: it violated the Same-Origin Policy.

If you visited a malicious page from any Apple device, it could read authenticated cross-origin data from other sites used, including access tokens and other sensitive information, making account takeovers trivial. The video below shows the proof of concept sent to Apple, demonstrating the leakage of sensitive data from both Apple Connect and iCloud/Apple ID endpoints. Although this demonstration focuses on Apple services, the issue affects all websites. This means that visiting a malicious website could risk exposing sensitive data from other domains.

The Sandbox Russian Doll

Browser exploitation in 2026 is very similar to being trapped in a Russian doll. You start in the smallest doll and every time you escape from a level, you are still trapped in another. Finding a low-level memory bug is not the same as finding an exploit. Most of these bugs die in the gap between "memory corruption occurred" and "something significant crossed a security boundary".

Outside is the browser process model. Even if the renderer code goes wrong, the browser tries to keep that damage within the web content process. Inside this, there is the web security model: Same-Origin Policy, CORS, opaque responses, cookie scope, and credential mode. Even if a page can trigger a cross-origin request, the renderer, and in particular the [Gigacage](https://webkit.org/blog/8300/click-to-flash-javascript-and-the-gigacage/), should prevent it from reading the response. That's the theory, at least.

The Exploit Chain

The exploit chain is simple:

• get an opaque response
• clone it
• read from the resulting readable stream

And that's it. Now we can read data we shouldn't be able to read. This is a serious problem because it means a malicious website could read sensitive data from another website. For example, a malicious website could read a user's access tokens from another website, allowing the attacker to access the user's account. Additionally, this bug could be used to bypass other security measures, such as origin checks.

The Impact

The impact of this bug is significant. An attacker could use this bug to steal sensitive data from other websites. For example, a malicious website could read a user's access tokens from another website, allowing the attacker to access the user's account. Additionally, this bug could be used to bypass other security measures, such as origin checks.

The Fix

The fix for this bug is simple: don't allow cloning of opaque responses. This is what WebKit did. Now, if you try to clone an opaque response, you will get an error. This prevents attackers from exploiting this bug.

This bug is an example of how even a small mistake can have serious consequences. It is important that developers are aware of these issues and that browsers are designed to prevent them. Additionally, it is crucial that users are aware of the risks associated with browsing unsafe websites.