Anthropic releases Claude Fable 5, a 'Mythos-class' AI model with safeguards

Anthropic has officially launched its latest AI model, Claude Fable 5, which is categorized as a 'Mythos-class' system. However, this release comes with significant safeguards designed to prevent potential misuse, particularly in sensitive areas like cybersecurity. In a statement, Anthropic highlighted that the enhanced safeguards were essential due to the model's capabilities. The company noted, "Without these measures, Fable 5's abilities could be exploited, leading to serious consequences in cybersecurity and other critical fields." As a result, when users make inquiries related to cybersecurity, biology, or chemistry, those requests will be redirected to Anthropic's Opus 4.8 model. Users will be notified when this rerouting occurs. Additionally, Anthropic is progressing with Project Glasswing, which was initially revealed in April. This initiative provided select tech, cybersecurity, and financial firms with early access to the first iteration of their Mythos series, known as Claude Mythos Preview. The company announced that a limited group of cyberdefenders and infrastructure providers would gain access to Claude Mythos 5, a variant of Fable 5 with some safeguards eased in certain areas. Anthropic claims that Mythos 5 possesses unparalleled cybersecurity capabilities, stating it is the most advanced model of its kind. The company plans to broaden access to Mythos 5 via a new trusted access program soon. Despite its limitations, Anthropic asserts that Fable 5 surpasses the performance metrics of any of its previous models. It has demonstrated exceptional capabilities across various benchmarks, including software engineering, knowledge work, and scientific research. The more complex the task, the more Fable 5 outperforms its predecessors. For instance, Stripe reported that Fable 5 was able to condense months of engineering work into just days. Furthermore, IMC noted that Fable 5 excelled in their trading analysis evaluations, showcasing its effectiveness in factual lookup and conceptual reasoning. Continuing the tradition of naming AI models after literary forms, Anthropic chose 'Fable' to reflect its connection to the Greek word 'mythos,' emphasizing storytelling in its design philosophy. This release is particularly timely for Anthropic, which recently filed a draft S-1, signaling its move toward a potentially lucrative IPO. This comes in the wake of similar announcements from its competitor, OpenAI, and the impending public offering of SpaceX’s xAI. The introduction of the Mythos model has raised eyebrows globally, with concerns about its potential to reveal vulnerabilities in software and cybersecurity. In response, President Trump recently signed an executive order allowing AI firms to voluntarily provide the federal government with access to advanced models ahead of their public release, while ensuring that mandatory reviews remain off the table.