Anthropic has officially launched its latest artificial intelligence model, Fable 5, tailored for general public use while imposing restrictions in certain sensitive areas. This model is the first offering from the company’s advanced Mythos class of AI technology, which was introduced earlier this year but remained accessible only to a limited group of partner institutions due to cybersecurity concerns.
Fable 5 aims to assist users in various tasks, including software code writing, debugging, addressing complex research inquiries, and image analysis. In conjunction with this launch, Anthropic has also made available an unrestricted variant known as Claude Mythos 5, but only for selected organizations that have prior access to this model family, particularly those engaged in cybersecurity through its Project Glasswing initiative. This initiative has seen the group of partners expand to approximately 200 organizations across over 15 countries, with plans for further growth.
Access to the Mythos models has been limited as a precaution against potential cybersecurity threats. The technology has the ability to swiftly identify vulnerabilities in critical infrastructures, such as banking systems and power grids. Initially, some critics labeled Anthropic’s warnings as exaggerated, suggesting they were aimed at garnering attention. However, subsequent evaluations by various companies—including those within the US government, which has a history of legal disputes with Anthropic—have confirmed the model’s capabilities. The White House has established a testing arrangement ensuring that powerful models from leading AI developers are evaluated before public release, with Mythos 5 being deployed in partnership with the US government.
To mitigate risks, Anthropic has planned that most queries related to sensitive areas like cybersecurity, biology, or chemistry directed at Fable 5 will be diverted to a less capable model, Opus 4.8, which was revealed to the public in late May. Additionally, the company has identified attempts to exploit its technology for training competing AI models in authoritarian regimes, leading to further limitations on Fable 5’s access.
In a proactive approach to safeguarding its technology, Anthropic hired external experts to engage in extensive red-teaming efforts, dedicating over 1,000 hours to exploring potential security vulnerabilities. The company also initiated a bug bounty program that incentivizes individuals to pinpoint security flaws; thus far, no one has successfully managed to bypass the specified restrictions.
The San Francisco-based startup faces its own challenges amid a contentious relationship with the Trump administration over its refusal to lift restrictions on mass surveillance and autonomous lethal weapons systems, resulting in the Pentagon terminating its contracts with the company, which previously held defense security clearance.
The pricing structure for Fable 5 is considerable, set at $10 per million input tokens and $50 per million output tokens—essentially double the cost associated with Opus 4.8. Given that tokens represent the units for pricing AI model usage, a programmer could exhaust one million tokens within a few hours during an intensive coding session.
Despite witnessing rapid revenue growth, Anthropic remains on a struggle toward profitability, largely due to high expenses associated with computing power. Recently, the company began a costly lease on a data center from Elon Musk’s xAI—part of SpaceX—costing $1.25 billion per month.
These developments coincide with a wave of financial excitement surrounding AI technologies, as both Anthropic and its competitor OpenAI have recently filed plans for initial public offerings (IPOs). Meanwhile, SpaceX is gearing up for what is expected to be a record-breaking market debut later this week.
