Anthropic's Opus 4.7: Safer AI for Coding

0:00It's April 16. From your city to your ears, this is California News today, powered by AI, and Thropic just rolled out Opus 4.7, an upgraded version of their top widely available AI model. This drop comes barely a week after they held back a more powerful one called Mythos. The new Opus shines in software engineering, tackling tough coding jobs that used to need heavy oversight, and it follows user instructions way better. While handling detailed high-res images like complex charts, they dial back some capabilities in Opus 4.7 on purpose, especially in cyber security, to make it safer than Mythos. During training, and Thropic even tested ways to reduce those cyber skills without hurting other features. Mythos proofs scary good at spotting and exploiting vulnerabilities in every major operating system and web browser when prompted. That's why they limited it to select businesses for security testing, sparking worries about AI risks in the wild. Now Opus 4.7 packs built-in safeguards that auto-block shady cyber requests.

1:05And Thropic plans to learn from real-world use to eventually unleash Mythos class models more broadly. All this heats up their rivalry with OpenAI, as in Thropic gained steam in AI coding tools for businesses and everyday users, even amid Pentagon talks, on safety.