Anthropic Introduced Claude Mythos 5 and Fable 5 with Cyberattack Protection

Anthropic unveiled two new versions of Claude: Mythos 5 for trusted organizations and Fable 5 for the general public. The second version was specifically developed with mechanisms that make it difficult to use in cyberattacks.

Division by Purpose

Anthhropic is taking the path of dividing its AI models depending on the target audience and potential risks. Claude Mythos 5 is intended for trusted partners, including cybersecurity companies, research organizations, and corporate clients who need the full capabilities of a language model without restrictions. Claude Fable 5 is released for public access. The company has embedded special restrictions in this version aimed at preventing the model's use in cyberattacks and other malicious purposes. This is a strategy that allows Anthropic to offer a powerful AI tool while minimizing direct responsibility for its potential misuse.

Protection Built Into Architecture

The key difference with Fable 5 is that the protection mechanism is not based simply on instructions or API-level restrictions (which are easy to circumvent), but on changes in the model's architecture and training process. Anthropic believes that this approach is more reliable and long-lasting than application-level control systems.

• Built-in restrictions in the model training process
• Architectural changes that prevent certain classes of tasks
• Public availability without licensing barriers, but with restrictions

Unlike a simple filter that catches malicious requests in real-time, built-in restrictions change the model's behavior at a fundamental level. This means that even when attempting to circumvent them, the system will be less useful for cyberattacks.

Growing Pressure on the Industry

Anthhropic's decision reflects growing pressure from regulators, politicians, and the public who demand that AI companies take responsibility for potential harm. Against the backdrop of debates about the possibility of using large language models in cyberattacks, the release of a 'safe' version looks like a strategic move. Other companies, including OpenAI and Google, are also under pressure and are developing their own control approaches. However, there is still no unified standard in the industry. Anthropic's approach with two-level division looks pragmatic: give trusted partners full access, while offering the public a limited but secure version.

What This Means

Anthhropic demonstrates that AI safety is beginning to be viewed as a separate product line rather than an add-on module. This could become a trend in the industry: instead of one model for all, companies will offer multiple versions with different levels of restrictions. However, questions remain about the practical effectiveness of built-in restrictions and whether users will accept limitations in exchange for a promise of security.