Can Anthropic Keep Its Exploit-Writing AI Out of the Wrong Hands?

Anthropic Mythos Preview can find and take advantage of zero-day security holes in popular web browsers and operating systems. Anthropic is giving Project Glasswing $100 million in Mythos usage credits and $4 million in cash to open-source security groups. The company says that Project Glass Wing could change the way cybersecurity works in a big way.

Mythos is interesting, but there is a warning: early Anthropic examples show promising vulnerabilities, but two examples are not enough to show patterns. "There's no way anyone can completely shield their data from cyber threats," says Melissa Ruzzi, director of AppOmni. Julian Totzek-Hallhuber from Veracode says, "What we can strive for is making it harder for attackers to gain access."

He says, "Until independent researchers with unrestricted access can run their own evaluations, it's best to stay skeptical." This is another result of the limited access model: claims can't be tested, so they stay unproven and can't be proven wrong. "Security Bosses Are All In on AI: Here's Why" is the title of the latest episode of ZeroOwl Confidential.

In this episode, Reddit CISO Frederick Lee and Omdia analyst Dave Gruber talk about how AI and machine learning are being used in Security Operations Centers (SOCs). They look at successful deployments and think about what the future holds for AI security products. Don't miss this episode, which is full of useful information!

Click here to learn more about ZeroOwls Confidential and how to get the most recent episode of the podcast. Go back to the page you were on. You can follow us on Twitter at @ZeroOwlConfidential and @ZOwlPolicymouth.

We have a page on Facebook as well.