Leaked documents expose Anthropic’s new AI model concerns
According to reports from Fortune, internal documents from Anthropic have been leaked, revealing details about a new generation of AI models called “Claude Mythos.” The leak happened because of a human error in Anthropic’s content management system configuration. Security researchers found that the company left nearly 3,000 unpublished assets in a publicly accessible data lake. These included images, PDFs, audio files, and draft blog posts.
What’s perhaps more concerning is that these documents suggest Claude Mythos is already in testing phase. The materials state that Anthropic believes this new model “poses unprecedented cybersecurity risks.” In a draft blog post, the company reportedly said the system is “currently far ahead of any other AI model in cyber capabilities.” They warned that it could lead to models that exploit vulnerabilities much faster than defenders can respond.
Anthropic’s cautious approach and past incidents
Because of these potential risks, Anthropic planned a careful rollout strategy. They wanted to give early access to cybersecurity defense organizations first. This would give defenders a head start in strengthening their systems against AI-driven attacks. The concern isn’t baseless either. Anthropic has previously reported that a Chinese state-sponsored group used Claude Code to infiltrate about 30 organizations, including tech companies and government agencies.
The leaked documents also mentioned an invite-only CEO summit planned at an 18th-century manor in the English countryside. Anthropic CEO Dario Amodei was set to host European business leaders there to discuss AI adoption and showcase unreleased Claude model capabilities.
Industry reactions and competitive landscape
Once the news spread on social media platform X, Elon Musk didn’t hesitate to comment. He wrote “Seriously troubling” and the post quickly gained tens of thousands of views and likes. Musk has a pattern of commenting on negative news about competitors, especially those he disagrees with. Anthropic was founded by former OpenAI employees, and Musk has been openly critical of both OpenAI and the broader AI industry’s approach to safety.
Meanwhile, Musk’s own AI company, xAI, recently launched a new paid subscription tier called “SuperGrok Lite” for $10 per month. They’ve placed limits on free users of Grok to push them toward paid plans. The company offers several subscription options, including SuperGrok for $30 a month, SuperGrok Heavy for $300 a month, and Grok Business for $30 per month.
I think this situation highlights the tension in the AI industry between rapid development and safety concerns. When companies are racing to develop more powerful models, security considerations sometimes take a back seat. The fact that these documents were accidentally made public shows how even basic security practices can be overlooked.
The human error in the CMS configuration is a reminder that technical systems are only as secure as the people managing them. Having thousands of unpublished assets publicly accessible seems like a basic oversight. It makes you wonder what other security gaps might exist in these fast-moving AI companies.
What’s interesting to me is how different companies approach these challenges. Anthropic appears to be taking a more cautious route with their new model, while others might push forward more aggressively. But then again, the leak itself suggests their internal processes might not be as tight as they’d like us to believe.
The industry reactions are telling too. Musk’s quick comment shows how competitive this space has become. Everyone’s watching everyone else, ready to point out flaws or missteps. It creates an environment where transparency might suffer because companies fear giving competitors ammunition.
I’m left wondering how much we don’t know about these AI models’ capabilities and risks. If this information was accidentally leaked, what other important details remain hidden? And how do we balance innovation with proper safeguards when the technology is advancing so quickly?
