← Retour au blog
tech 15 May 2026

Too Dangerous or Just Too Expensive? The Real Reason Anthropic is Hiding Mythos

Anthropic recently launched Claude Mythos Preview with strict distribution restrictions. But is it really due to its dangerous capabilities or just prohibitive costs?

Introduction

In April 2026, Anthropic quietly made waves with the launch of Claude Mythos Preview, an advanced AI model. Instead of making it widely accessible, Anthropic chose to limit access to about forty hand-picked organizations. This decision raises a fundamental question: is it for safety concerns or simply due to prohibitive costs?

Claude Mythos Preview: An Overview

Claude Mythos Preview is described as a model capable of discovering and exploiting zero-day vulnerabilities in real-time and at scale. This capability could potentially destabilize global digital infrastructure if misused. Anthropic claims the model has crossed a threshold never before reached by a commercial AI, justifying restricted release to give defenses a head start.

A Truly Dangerous Model?

The possibility that Mythos is too dangerous for wide release is a compelling argument. Indeed, exploiting zero-day vulnerabilities could wreak havoc on critical systems. However, it's crucial to remember that other AI models have been launched with similar, albeit less powerful, capabilities.

The Reality of Costs

Beyond safety, the computational costs associated with Mythos are astronomical. The infrastructure required to run such a model at scale is enormous, making wide release economically unviable. According to some estimates, computational costs for similar models have increased by 300% in recent years.

Anthropic's Strategy

Balancing these two explanations, a clear strategy from Anthropic emerges: limit access to optimize safety while managing costs. This approach also allows them to test the model in a controlled environment before any potential expansion.

Conclusion

The case of Claude Mythos Preview highlights the challenges AI companies face when developing cutting-edge models. The safety versus cost question is a dilemma that will continue to shape the industry.

Let's discuss your project in 15 minutes.

Anthropic Claude Mythos AI safety compute costs zero-day vulnerabilities
Deepthix newsletter · 100% AI · every Monday 8am

An AI agent reads tech for you.

Our AI agent scans ~200 sources per week and ships the best articles to your inbox Monday 8am. Free. One click to unsubscribe.

Visit the newsletter page →

Want to automate your operations?

Let's talk about your project in 15 minutes.

Book a call