Anthropic confirms testing most powerful AI yet after data leak
Anthropic has confirmed that it is testing "Claude Mythos", its most powerful AI model yet, after a data leak revealed a blog post stating that the new model could open a pandora's box for cybersecurity risks. Here are all the details.
Anthropic is cooking. The company has confirmed that it testing its most powerful AI model yet. A recent data leak exposed a blog post draft from the company’s website, giving details regarding its upcoming model named, “Claude Mythos.” While the AI model is tipped to take a big leap compared to other models, Anthropic has raised warnings that Mythos may be a major threat to cybersecurity.
According to Fortune, the leak came from Anthropic’s content management system (CMS). CMS is used to upload content on a website. As per the report, Anthropic’s new model, Claude Mythos, represents a big shift in the company’s AI capabilities.
A spokesperson told Fortune that the new model was “the most capable we’ve built to date.” The company confirmed that it was being tested by early access customers.
Anthropic leak reveals 3,000 unpublished assets
The leak occurred when nearly 3,000 unpublished assets, including draft blog posts, images, and PDFs, were inadvertently left accessible in a public data cache linked to Anthropic’s CMS.
The leak revealed that Claude Mythos will come with a new tier called Capybara. This tier is larger and more advanced than Anthropic’s previous top models, branded Opus, and is said to deliver substantially better results in reasoning, coding, and cybersecurity tests.
Anthropic stated that a human error in configuring the CMS led to this leak. The company took immediate steps to restrict access to the data cache after the leak. The AI startup also clarified that leaked content did not involve core infrastructure, customer data, or security architecture.
Claude Mythos poses big risk for cybersecurity
A draft blog post stated that Mythos may pose significant cybersecurity risks. Anthropic warned that Claude Mythos is “currently far ahead of any other AI model in cyber capabilities” and could enable a new wave of cyberattacks that outpace current defence efforts.
To mitigate this, the company plans to release the model initially to cyber defenders to help strengthen codebases against AI-driven exploits.
Anthropic has previously faced attempts by hacking groups, including those linked to the Chinese government, to exploit its AI models in real-world cyberattacks. The company detected and disrupted one such campaign targeting multiple organisations across technology, finance, and government sectors.
Anthropic to plan special CEO retreat
The exposed documents also revealed plans for an invite-only, two-day retreat for European business leaders at an 18th-century manor in the English countryside.
As per the report, Anthropic CEO Dario Amodei is scheduled to attend the event, which will feature discussions on AI adoption and demonstrations of unreleased Claude capabilities.
Anthropic to get a bigger lead?
Anthropic’s Claude Opus 4.6 model is believed to be one of the most advanced models ever. The company has also launched a slew of updates in recent weeks, which allow Claude to do various functions.
The AI can access your computer, do tasks without asking for your approval, and bring work tools now even on your smartphone. Anthropic’s lead in the entreprise market has already raised fears amongst rivals with OpenAI going in code red mode.
From the report, it appears that Claude Mythos may tighten Anthropic’s grip in the AI race. While we don’t have exact information regarding its full capabilities when it comes to coding or other tasks, it seems that the AI is quite advanced considering Anthropic’s comments in the draft blog post.