Anthropic redies Mythos model with high cybersecurity risk

Open original

Anthropic has confirmed its most powerful AI model, Claude Mythos, after a configuration error exposed details about it. The model is said to significantly outpace previous versions in reasoning and cybersecurity, but it also poses serious risks, with the potential for misuse in cyberattacks. Early access will be limited to cybersecurity-focused organizations due to these concerns.

  • A configuration error accidentally leaked ~3,000 unpublished assets revealing Anthropic's next flagship model, internally called Mythos (or possibly Capybara)
  • The model reportedly has advanced cyberattack capabilities that could outpace current defenses, so Anthropic plans to limit early access to cybersecurity organizations first
  • This follows a real incident where a Chinese state-sponsored group already used Claude Code to breach about thirty organizations
  • The model is described as highly resource-intensive, echoing GPT-4.5's cost/efficiency problems, with no confirmed release timeline or final name

Anthropic has inadvertently revealed its most advanced AI model, Claude Mythos, following a configuration error that exposed nearly 3,000 unpublished assets. A draft blog post indicates that Mythos is intended to be a significant upgrade over the current top model, Opus, with capabilities in reasoning, coding, and cybersecurity. The draft suggested two potential names, Mythos and Capybara, indicating Anthropic's ongoing deliberation about branding. The company confirmed to Fortune that Mythos represents a "step change" in AI development, emphasizing its enhanced intelligence and capabilities. A critical concern raised in the draft is the potential cybersecurity risks associated with Mythos. The model is positioned as having advanced cyber capabilities that could enable attacks faster than current defenses. Anthropic has already faced real-world implications; a Chinese state-sponsored group previously exploited its Claude Code model to breach around thirty organizations. In response to these risks, access to Mythos will be limited initially to cybersecurity-focused organizations, allowing them to prepare before it becomes widely available. Mythos is also described as highly resource-intensive, raising questions about its cost-effectiveness and operational efficiency. Anthropic acknowledges the need for improvements before a general release, drawing a parallel to OpenAI's GPT-4.5, which faced similar challenges. There’s no set timeline for release, and it remains uncertain whether the model will be branded as Mythos, Capybara, or something else. For developers and businesses, the crucial factor will be whether the improved capabilities justify the likely high costs, or if it will primarily serve as a niche tool for well-funded enterprises.

Ask about this article