Anthropic has announced the release of its highly anticipated AI model, known as Mythos, cautioning that it could pose significant risks if misused. Nicole Nguyen from The Wall Street Journal offers insights into these dangers and presents essential cybersecurity advice for everyday tech users.
The new model, dubbed Claude Fable 5, is designed with built-in safeguards to prevent misuse in critical fields such as cybersecurity and bioweapon research. While it enables users to interact with Mythos, the model will redirect inquiries about sensitive topics, such as bioweapons or software vulnerabilities, to the older Opus 4.8 version of the Claude chatbot.
Although Fable 5 will be priced higher than Opus 4.8, it boasts improved retention capabilities, enhancing its ability to tackle large and complex tasks with reduced user input. Dianne Penn, head of product management, research, and labs at Anthropic, emphasized the goal of delivering advanced intelligence to users while ensuring safety.
Despite these safety measures, the potential for hackers to manipulate Mythos through a practice known as jailbreaking remains a concern. Anthropic has conducted extensive testing to mitigate this risk, but continued diligence will be necessary.
Since its unveiling in April, Mythos has prompted the Trump administration to reassess its artificial intelligence strategy and raised alarms among cybersecurity experts, who warn that advanced AI models like Mythos are creating a surge in software bug reports.
Anthropic has so far limited full access to the Mythos model, currently available to around 200 organizations, including major technology firms like Verizon and Microsoft. These companies are actively conducting security tests on their systems and products in anticipation of potential threats from similar technologies.
Additionally, Anthropic is upgrading access for its early partners from Mythos Preview to Mythos 5, with plans to gradually expand access through a structured trusted-access program in the future.
