AI Model Uncovers Decades-Old App Vulnerabilities

0:00Anthropic just dropped a bombshell with their upcoming AI model, Claude Mythos. This thing sniffed out thousands of vulnerabilities in everyday apps, some without any fixes available. The oldest one dates back 27 years, and none were spotted by the original developers until now. To tackle this, Anthropic launched a project called Glasswing, teaming up with about 40 organizations that include cyber security heavyweights like CrowdStrike and Palo Alto Networks, plus tech giants Amazon, Apple, Microsoft, Cisco, Broadcom, and even. The Linux Foundation. Developers are stunned because these flaws are super subtle. Mythos caught one in video software that got tested over 5 million times without anyone. Noticing, it's a wake-up call on how AI can outpace human checks in spotting hidden risks. Mythos isn't even built as a bug hunter, it's a general purpose model. A recent cold leak had an anthropic warning about major cyber security threats, so they're sharing it cautiously with partners to stay ahead.

1:01Meanwhile, they're talking with the US government, even as a White House order to cut ties sits on hold in court. This could reshape how we defend software before the next big threats hit. From your city, powered by AI, this is Canada News Today.