News

This isn’t about making AI sound smarter. It’s about making it more useful.

For the first time, the US government is stepping in before some of those models are publicly released.

That shift matters. For years, the AI industry competed over who could write better text, answer faster or generate more eye-catching images. Now the battle is moving somewhere else: which model can actually help users complete valuable tasks with fewer mistakes and less supervision.

The tool is developed by Anthropic Labs — the company's experimentation arm — and is powered by Claude Opus 4.7, Anthropic's most capable publicly available model, launched the same day.

The trick developers were using

That decision is not framed as vague AI-doom messaging. In its own materials, the company says Claude Mythos Preview has already uncovered thousands of severe vulnerabilities, including bugs affecting major operating systems, browsers and widely used software components. Anthropic also says the model showed an unusual ability to reason about exploit chains and, in some cases, develop attack paths with far more autonomy than previous Claude models.

That technical detail was enough for developers and researchers to reconstruct parts of the system’s internal source code. What matters here is not just that code was exposed, but what kind of code: the operational layer that turns Claude into a real-world coding assistant capable of interacting with files, executing tasks and working with external tools.

The premise is concrete: build real products using Anthropic's AI models — that is, Claude — in a single day, with direct mentoring from Kaszek, Anthropic and Digital House teams. Every registered participant will receive Anthropic API credits to use during the event at no cost.