Anthropic will not release its Claude Mythos Preview model to the general public. The startup cited the model's advanced ability to identify and exploit software vulnerabilities.
Testing by Apple, Google, and JP Morgan revealed capabilities that could be dangerous in the hands of malicious actors. Anthropic will instead share the tool exclusively with these major corporations to bolster critical internet infrastructure security.
The decision highlights the evolving cybersecurity risks associated with sophisticated AI models. This move underscores the growing tension between advancing AI capabilities and ensuring safe deployment for the technology sector and its investors.