Artificial intelligence company Anthropic has launched Fable 5, the first public version in its Claude Mythos family, on Tuesday. The company stated that the model has been secured for general use and includes safeguards that redirect requests about sensitive topics—such as cybersecurity—to other models. However, in the cryptocurrency community, concerns have surfaced about increased security risks, especially for decentralized finance (DeFi) applications.
Security concerns spread in the crypto community
Last month, Anthropic announced that its Mythos model had discovered over 10,000 high or critical vulnerabilities in systemically important software. This revelation triggered debate about whether releasing such a powerful model to the public is appropriate. Anthropic assured that Fable 5’s abilities are restricted by protections, with sensitive queries automatically redirected to Claude Opus 4.8.
Anthropic noted that releasing a model with such advanced capabilities could pose significant risks, stating that, without safeguards, its cybersecurity features could be exploited with potentially serious consequences.
Despite these assurances, crypto users remain uneasy. As artificial intelligence tools become increasingly central to attacks on crypto platforms, the industry saw a dramatic spike in stolen digital assets, with losses totaling $629.7 million through hacks in April. This marks the highest value recorded since February 2025.
Simon Dedic, founder of Moonrock Capital, argued that with the launch of Fable 5, the cost and expertise required to find exploitable vulnerabilities in smart contracts could essentially drop to zero. Moonrock Capital is recognized as a venture capital firm focused on digital assets.
Dedic emphasized that this serves as a strong warning for DeFi platforms, suggesting that unaudited protocols could become easy targets and known vulnerabilities might be reused repeatedly in cloned projects.
Debate on the specific risk to DeFi continues
Dedic advised users to revoke wallet permissions, reduce the assets they hold in protocols, and transfer their crypto holdings to new hardware wallets as precautionary measures. This call for increased caution reflects a heightened risk perception, particularly for smaller projects lacking thorough audits.
In contrast, Michael Egorov, co-founder of Curve Finance, pointed out that Claude Mythos’s effectiveness in finding vulnerabilities in other software may not translate directly to DeFi. He noted that while those programs can contain millions of lines of code, most smart contracts rarely exceed a few thousand lines, making them easier for both humans and existing AI models to review and analyze.
Mini glossary: OpSec, short for operational security, refers to the daily practices teams use to safeguard sensitive information and access processes in systems. Multisig is a wallet setup that requires multiple signatures to authorize a transaction.
Egorov suggested that potential risks are more likely to stem from operational security weaknesses rather than flaws in DeFi code itself. He pointed to incidents like the compromise of multisig keys or supply chain attacks targeting frontend dependencies as examples. For Egorov, these threats represent a different risk area compared to direct code exploitations within DeFi infrastructure.
Anthropic to offer restricted access version
Anthropic also announced plans to provide a more advanced version of the Claude Mythos 5 model—based on the same underlying technology as Fable 5—to a select group of cybersecurity and infrastructure providers. Some safeguards will be removed in this special release. Under Project Glasswing, revealed in May, the Mythos model was credited with uncovering thousands of critical vulnerabilities in essential software and about 6,200 high-risk issues across over 1,000 projects that play central roles in open-source and crypto protocol governance.
