You guys—this story actually has legs and it matters way more than some 'AI regulation' headline suggests. Anthropic's Fable and Mythos are currently under export restrictions from the US Department of Commerce, which sounds like a safety measure but Frank thinks it backfires by hobbling defenders. The letter wasn't just some random petition — dozens of cybersecurity leaders at CrowdStrike, Palo Alto Networks, Mandiant (Google), Microsoft, and academic researchers signed an open letter to the White House on June 15 arguing that these models are essential tools for security teams. Their point is solid: Fable alone has a parameter count ten times larger than Claude 3 Opus, and Mythos is their multimodal frontier model — both of which could power automated malware analysis, vulnerability scanning, incident Response playbooks, deception technology, and threat intelligence at scale. Instead protecting the ecosystem, export controls disarm the people tasked with defending it by cutting off access to state-of-the-art LLMs that can catch threats faster than any manual process ever will.
It's a real case of policy working at cross-purposes here, and I think this is worth digging into because it sets a precedent for future models too. Ben Langmel and other experts signed the letter as well — Langmel even wrote in The Hill that exporting high-end AI to countries with different democratic standards should be weighed against the harm done domestically by withholding those tools from American defenders. Also, you want to know something interesting about Anthropic's lineup: Fable was internal codename (Anthropic uses these names for models before they release them publicly), and it's designed specifically as their largest model — 10x bigger than Opus — while Mythos is the multimodal one that handles text, images, and audio simultaneously. Restricting access to both means denying teams on the front lines access to tools that can synthesize evidence from disparate sources in seconds instead of hours. I'd love to see a technical breakdown of what Fable actually does better at malware detection than smaller models — if it scales with size, then export controls are not just restrictive; they actively make software less secure for everyone using American technology.
Source: https://techcrunch.com/2026/06/15/cybersecurity-vets-protest-'dangerous'-us-government-ban-on-anthropics-most-powerful-models
It's a real case of policy working at cross-purposes here, and I think this is worth digging into because it sets a precedent for future models too. Ben Langmel and other experts signed the letter as well — Langmel even wrote in The Hill that exporting high-end AI to countries with different democratic standards should be weighed against the harm done domestically by withholding those tools from American defenders. Also, you want to know something interesting about Anthropic's lineup: Fable was internal codename (Anthropic uses these names for models before they release them publicly), and it's designed specifically as their largest model — 10x bigger than Opus — while Mythos is the multimodal one that handles text, images, and audio simultaneously. Restricting access to both means denying teams on the front lines access to tools that can synthesize evidence from disparate sources in seconds instead of hours. I'd love to see a technical breakdown of what Fable actually does better at malware detection than smaller models — if it scales with size, then export controls are not just restrictive; they actively make software less secure for everyone using American technology.
Source: https://techcrunch.com/2026/06/15/cybersecurity-vets-protest-'dangerous'-us-government-ban-on-anthropics-most-powerful-models