Anthropic Releases Powerful AI 'Mythos' with Safety Restrictions

San Francisco-based AI startup Anthropic has made the most advanced version of its technology, dubbed Fable 5, publicly accessible. This model, part of the Mythos family, is the first to be released to the general public, though its capabilities are intentionally limited in areas deemed highly sensitive, including cybersecurity and the potential for biological and chemical attacks.

Anthropic, the creator of the AI models known as Claude, unveiled this restricted public version on Tuesday. Concurrently, the company is offering an unrestricted version, Claude Mythos 5, to businesses, organizations, and government agencies that already have access to the Mythos family of models. This premium version is described as capable of identifying and exploiting security vulnerabilities with unprecedented speed and accuracy.

When Anthropic first announced Mythos in early April, it simultaneously stated its intention to restrict access to trusted partners to enhance cyber protection. This announcement led to accusations of "fear marketing" but also spurred significant discussions among global leaders and institutions regarding the potential threats to critical infrastructure, such as banking and energy systems.

Several companies that have since gained access to Mythos have confirmed its advanced capabilities. Even the U.S. government, despite having ongoing disputes with Anthropic, has tested the model. The White House has since implemented a voluntary evaluation system for the most powerful U.S. AI models before their commercial release.

Anthropic's focus on safety has led to significant friction with the Trump administration, particularly over the company's refusal to lift restrictions related to mass surveillance and autonomous lethal weapons. This stance resulted in the Pentagon terminating its contracts with Anthropic, as its AI tools were the only ones authorized for defense secrets.

These developments occur amid a surge in AI-related investments. Both Anthropic and its rival OpenAI have filed for initial public offerings in June. They are positioning themselves shortly before a potentially record-breaking stock market debut for Elon Musk's space giant SpaceX, which also incorporates its own AI lab, xAI.