Anthropic’s cybersecurity AI model that is restricted, Mythos, faced a turbulent week, as OpenAI Chief Executive Sam Altman publicly dismissed the company’s safety stance and Bloomberg stated that unauthorised users were able to access the model through an outside vendor.
Altman who was speaking in the show Core Memory, characterised Anthropic’s decision to keep Mythos from releasing to the public as fear-based advertising — which he said was designed to put AI within the reach of the most privileged. He suggested that branding the product as incredibly powerful was a successful marketing strategy, even though some critics pointed out it was Altman himself has used an existential AI rhetoric to highlight the work of OpenAI.
Anthropic has announced Mythos earlier in the month, restricting access to around 40 organizations under a program called Project Glasswing The company cited fears that the model’s cyber-attack capabilities could be exploited by criminals. Apple was one of the chosen recipients.
However, despite these precautions Bloomberg revealed that users of a private online forum were able to access Mythos at the time of the announcement, according to reports that they were able to determine the location of the model based on Anthropic’s well-known URL conventions and also via access granted to an employee of an external contractor. Anthropic confirmed that it was looking into the allegations, but claimed it could not find any evidence of an impact on its systems.
Additionally, OpenAI launched ChatGPT Images 2.0 which is a vastly improved image generation model that incorporates integrated web search Multi-panel comic creation enhanced text rendering for non-Latin characters, and outputs with up to 2K resolution. The model’s design is not known, but OpenAI confirmed that it has reasoning capabilities that permit it to check and refine the output of its own images.
