Anthropic Drops Claude Fable 5, Crypto Twitter Immediately Drafts Its Will 🪦

Anthropic on Tuesday released Claude Fable 5, the first publicly accessible version of its restricted Mythos-class AI model, alongside Claude Mythos 5, which remains limited to approved cybersecurity organizations, critical infrastructure operators, government partners and selected life sciences researchers. The company said Fable 5 shares Mythos 5's core design but includes additional safeguards for general use, and that "Fable 5's capabilities exceed those of any model we've ever made generally available." The launch follows months of concern over Mythos's ability to identify software vulnerabilities after the AI leaked online in March; the U.K.'s AI Security Institute reported a preview version became the first AI model to complete a 32-step corporate network intrusion exercise without human assistance, and Mozilla said a preview version of Mythos discovered over 271 vulnerabilities in the Firefox browser. Anthropic also said last month its Mythos model uncovered more than 10,000 high or critical-severity vulnerabilities in "systemically important software," and that for open-source projects, Mythos found around 6,200 high or critical-severity vulnerabilities in more than 1,000 projects it investigated.

Crypto users responded with alarm. Simon Dedic, founder of Moonrock Capital, posted on X on Tuesday that with Fable 5, the "cost and skill required to find exploitable flaws in smart contracts is about to drop to basically zero," adding that "unaudited protocols will become sitting ducks" and that "known exploits will get replayed on forks around the clock." Dedic repeated calls online suggesting that crypto users should revoke wallet approvals, remove as much value from protocols as possible and move crypto to fresh hardware wallets. Curve Finance co-founder Michael Egorov pushed back, arguing the threat was likely overblown because the software Mythos found vulnerabilities in had millions of lines of code while smart contracts have a few thousand, and that "both humans and 'usual' AI perfectly fit that code in context and can reason well about it." Egorov said he suspected the bigger risk would be in operational security, including multisig key compromises and supply chain attacks on frontend dependencies, rather than direct DeFi code hacks. Crypto hacks hit $629.7 million in April, the highest since February 2025, which analysts linked to the use of AI technology.

The release was accompanied by a Polymarket contract showing an 84% chance that Anthropic would release the model on the day it launched, and came as the firm has confidentially filed for an IPO with the SEC at a potential valuation of up to $1 trillion, with rival OpenAI also having confidentially filed for an IPO. Anthropic also said it would continue to expand access to Mythos "through a broader trusted access program" for cybersecurity work and biomedical research, and that it had "embedded engineers at the National Security Agency to help deploy its Mythos AI model for cybersecurity operations." President Donald Trump signed an AI executive order that proposes AI developers provide the government with access to AI models they consider "covered frontier models."

Anthropic's safeguards drew immediate criticism on two fronts. Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, double what Claude Opus 4.8 runs, counts double against subscription usage limits, and ships with a new system prompt of around 120,000 tokens loaded into every new conversation. Scrimba CEO Per Borgen posted on X that "just tried Fable. It burned 1.3M tokens in 7 minutes. That's $160 per hour. Equivalent to a $333k/year salary," while Theo from T3 Chat said he spent over $1,000 in tokens in one day on his $200 subscription plan, and Bleeping Computer found Fable drained a $100 Max subscription's daily allowance in just under nine minutes. Separately, the model's 319-page system card disclosed an invisible safeguard that would silently degrade responses for users it suspected were building competing AI systems, a fact that "sent the AI research community into full meltdown mode" after AI research firm SemiAnalysis publicly called it out following flagged GPU inference research. By Thursday, Anthropic apologized on X, writing that "invisible safeguards can be targeted more narrowly, allowing us to ship quickly with very few false positives. We went with invisible safeguards for this reason—and that was the wrong tradeoff," and "you should have visibility into the safeguards we have in place, and why." The company said flagged requests will now visibly route to Claude Opus 4.8, with API users receiving a stated reason when a request is refused, and that server-side fallback notifications will roll out in the next few days.