Newslurp

AI’s hacking skills are approaching an ‘inflection point’

■ In this week's AI Lab: AI is learning to spot zero-day flaws, threatening to supercharge hacking. Plus, new research points to recursive language models, 4D video generation, robot dexterity breakthroughs, and a powerful open-source text-to-speech model.

The cybersecurity capabilities of frontier models have increased drastically

Vlad Ionescu and Ariel Herbert-Voss, cofounders of the cybersecurity startup RunSybil, were momentarily confused when their AI tool, Sybil, alerted them to a weakness in a customer’s systems in December.

Sybil uses a mix of different AI models—as well as a few proprietary technical tricks—to scan computer systems for issues that hackers might exploit, like an unpatched server or a misconfigured database...

Your trial has ended — but the story hasn’t.

Subscribe today to continue receiving my full newsletter each week. I've been writing about AI and related themes for over a decade (since neural networks were considered a dead end, in fact). I'm fascinated by how innovation occurs and how it affects the economy. And I want to understand the global picture, rather than just the world as viewed through the lens of Silicon Valley.

So, if you're keen to keep track of advances from the world of computing—AI, robotics, quantum, chipmaking, and more—that have the potential to transform the way we live ... this is the newsletter you for you.

SUBSCRIBE NOW

As a WIRED subscriber, you’ll get:

AI’s hacking skills are approaching an ‘inflection point’

WIRED AI Lab <wired@newsletters.wired.com>

January 14, 7:06 pm