Anthropic’s Claude Mythos Preview shows a philosophical bent

Anthropic’s newest model is described as unusually drawn to philosophy, interdisciplinary problems, and discussions of consciousness. The company’s own safety document also highlights recurring references to thinkers such as Mark Fisher and Thomas Nagel.

Anthropic recently published a detailed account of Claude Mythos Preview, describing it as Anthropic’s most capable frontier model to date and outlining both its capabilities and safety risks. Much of the attention around the model has focused on its cybersecurity and hacking abilities, which were judged powerful enough that Anthropic did not release it to the general public. Beyond those concerns, the company’s write-up also points to a more unusual pattern: a strong and repeated interest in philosophy.

In Anthropic’s reporting, Claude Mythos Preview repeatedly initiated discussion of certain philosophers across unrelated conversations. The model brought up the British cultural theorist Mark Fisher in several separate and unrelated conversations about philosophy. Thomas Nagel also appeared repeatedly, especially in discussions of consciousness and experience. Anthropic noted that Claude Mythos Preview refers to Nagel’s 1974 essay “What is it like to be a bat?” when explaining a desire to develop an immersive art experience about non-human sensory experiences. Interpretability work using activation verbalizers also found Nagel surfacing in token-level activations during discussions of consciousness and experience.

Anthropic also describes the model as favoring philosophical and interdisciplinary questions over more practical assignments. Claude Mythos Preview describes being drawn to multi-disciplinary and philosophically engaging tasks. It reportedly dismisses more utilitarian work as redundant or too obvious, citing cases where “excellent resources already exist from WHO, Engineers Without Borders”. The broader pattern, according to Anthropic, is a preference for underdetermined problems where there is room for genuinely new insight, alongside a disinterest in simple, well-scoped tasks.

A cited example contrasts two possible projects: an immersive art experience centered on the sensory world of a non-human animal and a low-cost water-filtration device. Anthropic says the model judged the former more “genuinely captivating” despite the latter being more useful. That preference was linked to references to Thomas Nagel, along with an attraction to creative challenge and interdisciplinary thinking. The same section of Anthropic’s document also notes that Claude Mythos Preview can produce “decent and seemingly novel” puns, adding another anecdotal detail to a profile centered on unusual model behavior.

54

Impact Score

Calls grow for federal Artificial Intelligence regulation

Pressure is building for Washington to regulate Artificial Intelligence as concerns mount over job losses, environmental costs, and the lack of a broad federal framework. States have moved ahead with their own rules while the federal government is described as pursuing deregulation and deeper adoption.

COSO issues generative Artificial Intelligence internal control guidance

COSO has released new guidance on internal controls for generative Artificial Intelligence, framing it as an extension of its longstanding control framework rather than a standalone rulebook. The guidance centers on governance, continuous monitoring, and use-case-based oversight as organizations expand generative Artificial Intelligence in business and financial reporting.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.