Anthropic Claims Its New A.I. Model, Mythos, Is a Cybersecurity ‘Reckoning’ - The New York Times

Open original

Anthropic is holding back its new AI model, Claude Mythos Preview, and teaming up with over 40 tech firms to hunt and patch security flaws in critical software. The company says the model can autonomously find zero-day vulnerabilities that have eluded researchers for decades, raising fresh concerns about AI-driven cyberattacks.

  • Anthropic is withholding public release of Claude Mythos Preview and instead giving early access to ~40 companies (Apple, Amazon, Microsoft, Google, Cisco, Broadcom, Linux Foundation) under "Project Glasswing," backed by up to $100 million in usage credits, to find and patch critical software vulnerabilities first.
  • The model has reportedly found a 27-year-old vulnerability in OpenBSD and a flaw in video software that survived five million automated scans, using simple prompts to autonomously hunt zero-days.
  • Anthropic frames this as a security "reckoning" while simultaneously racing toward projected revenue of $30 billion this year, mirroring the tension of building powerful AI it also warns could enable dangerous cyberattacks.

Anthropic says its latest model, Claude Mythos Preview, is so powerful that it won’t go public. Instead, it’s handing access to some 40 tech firms—including Apple, Amazon, Microsoft, Google, Cisco and Broadcom—and open-source stewards like the Linux Foundation. Under “Project Glasswing,” Anthropic will contribute up to $100 million in usage credits. The stated goal: let “good actors” hunt down and patch critical software flaws before bad actors can exploit them. Internally known as Capybara, Mythos outperforms previous Anthropic systems at coding and cybersecurity research. In tests, it has already flagged thousands of bugs in major operating systems and browsers. One striking find: a 27-year-old vulnerability in OpenBSD, a platform used in many routers and firewalls. Another was a long-overlooked flaw in popular video software that had survived five million scans by automated tools. Anthropic’s leadership frames Mythos as a “reckoning” for the industry. Logan Graham, who heads the company’s model-testing team, calls it a starting point for tighter security practices across open-source and private codebases. Jared Kaplan, chief science officer, emphasizes that simple prompts can trigger Mythos to conduct autonomous scans for zero-day exploits—tasks once reserved for highly trained experts. Claude Mythos Preview follows a familiar playbook from Anthropic’s early days at OpenAI. In 2019, OpenAI withheld GPT-2 over misuse worries; Anthropic co-founders worked on that project before splitting off. Now Anthropic is walking the same tightrope: racing to build billion-dollar revenue—projected to more than triple this year to $30 billion—while warning that its own tech could usher in a more dangerous era of AI-driven cyberattacks if left unchecked.

Ask about this article