AI & Technology
Unreleased Anthropic AI Turns Security Vulnerabilities Into Exploit Code
10/29/2026
Model mapped flaws in browsers and operating systems, sending shockwaves through the tech industry
An unreleased artificial intelligence (AI) system from Anthropic managed to identify security vulnerabilities in widely used environments — including Chrome, Safari, Windows, and Linux — and chained those flaws together to create functional exploit code. The disclosure was made by the Frontier Red Team, the company's security division, in a report published this Tuesday (27).
Specialists explained that the attack code was built in a controlled lab setting following initial prompts given to the model, with no record of real-world attacks exploiting these flaws on the internet. Among the examples cited by researchers is an experiment in which the technology launched the calculator app on a Mac computer running the Chrome browser immediately after a webpage loaded — with no additional clicks required from the user.
Source: Estadão ↗
1 comentário
Comentários
- Vinícius Graff29/10/2026Ta tenso pessoal! 😲