Anthropic's Mythos AI Achieves 72.4% Success Rate in Generating Browser Sandbox Exploits
By
jnord
Sesame, salt, and substance. A flagship bake.
Summary
Anthropic's Mythos research preview demonstrates a significant advancement in AI's ability to generate working exploits for browser sandboxes, achieving a 72.4% success rate compared to under 1% just months ago. This breakthrough challenges the fundamental security model of the modern internet, which relies on sandboxes to contain untrusted code from JavaScript, cloud VMs, and ad iframes. The article examines the implications for cybersecurity and the trajectory of frontier AI models in breaking the foundational security assumptions that have protected internet users for nearly two decades.
Key quotes
· 4 pulledFor nearly 20 years the deal has been simple: you click a link, arbitrary code runs on your device, and a stack of sandboxes keeps that code from doing anything nasty.
Anthropic just shipped a research preview that generates working exploits for one of them 72.4% of the time, up from under 1% a few months ago.
Browser sandboxes for untrusted JavaScript, VM sandboxes for multi-tenant cloud, ad iframes so banner creatives can't take over your phone or laptop - the modern internet is built on the assumption that those sandboxes hold.
That deal might be breaking.
You might also wanna read
Google reports first evidence of hackers using AI to develop zero-day security exploit
Google has reported evidence of hackers using AI to develop a zero-day security vulnerability, marking the first time the company has observ
politico.com·10h ago
AI-Powered Bug Discovery Finds 271 Hidden Vulnerabilities in Firefox, Signaling New Era for Software Security
Security Now episode 1080 analyzed how frontier AI models (specifically Claude) discovered 271 hidden bugs in Firefox's codebase, as documen
AI-Assisted Exploit Development Time Drops from 125 Days to 12 Hours, Outpacing Scanners
New research from Cogent Research analyzing 69,159 CVEs reveals that AI-assisted attackers have reduced exploit development time from 125.3
Anthropic's Mythos cybersecurity AI model accessed by unauthorized users via third-party contractor
Anthropic's powerful Mythos cybersecurity AI model, described as potentially dangerous in the wrong hands, was accessed by unauthorized user
The Verge·1mo ago
Anthropic's Claude Mythos AI model accessed by unauthorized users despite security claims
Anthropic's tightly controlled rollout of its Claude Mythos AI model, touted as too dangerous for public release due to its advanced cyberse
The Verge·1mo ago
Google expands CodeMender AI security tool access, competing with Anthropic's Mythos
Google is expanding access to CodeMender, an AI-powered code security tool originally debuted in October 2024. At I/O, the company announced
The Verge·12d ago