Artificial intelligence company Anthropic launched Fable 5, the first public version of the Claude Mythos family, on Tuesday. The company announced that the model has been made safe for general use and includes protections that direct requests to a different model in some sensitive topics such as cyber security. On the other hand, concerns have come to the fore in the cryptocurrency community that security risks may increase, especially in decentralized finance applications.
Security concerns came to the fore in the crypto community
Anthropic announced last month that its Mythos model had detected more than 10,000 high or critical severity vulnerabilities in systemically critical software. This statement accelerated the debate about whether it is right to publish such a model publicly. The company stated that Fable 5 was limited by protections and that some risky titles were directed to Claude Opus 4.8.
Anthropic stated that releasing a model with this level of capability carries risks, and that without protections, capabilities, especially in the field of cyber security, can be misused to cause serious harm.
These assurances did not completely assuage crypto users’ concerns. While it is noted that artificial intelligence is increasingly used in attacks against crypto platforms, the total value of crypto assets stolen through attacks in April reached 629.7 million dollars. This amount was recorded as the highest level since February 2025.
Moonrock Capital founder Simon Dedic argued that with Fable 5, the cost and expertise required to find exploitable vulnerabilities in smart contracts could drop to almost zero. Moonrock Capital is known as a digital asset-focused venture capital firm.
Simon Dedic emphasized that this is a strong warning for DeFi, that unaudited protocols can become easy targets and known vulnerabilities can be constantly reused in forked projects.
Threat assessment for DeFi remains controversial
Dedic suggested that users revoke their wallet permissions as a precaution, reduce the amount of assets they hold in protocols, and move their crypto assets to new hardware wallets. This call pointed out that the perception of risk has increased, especially for small-scale projects with weak inspection processes.
On the other hand, Curve Finance co-founder Michael Egorov said that Claude Mythos’ success in finding vulnerabilities in other software may not yield the same result on the DeFi side. According to Egorov, while the software in question contains millions of lines of code, smart contracts are often limited to a few thousand lines, and this structure can be analyzed more easily by both humans and existing artificial intelligence systems.
Mini dictionary: OpSec, or operational security, refers to practices aimed at protecting sensitive information and access processes in the daily working order of systems and teams. Multisig is a wallet structure that requires the signature of more than one key to confirm a transaction.
Egorov noted that the potential risk can be seen in operational vulnerabilities rather than directly in the DeFi codebase. He cited breaches similar to the compromise of multisig keys and supply chain attacks targeting front-end dependencies as examples of this. According to him, such attacks pose a different risk area compared to direct code exploits of pure DeFi architecture.
Anthropic will also offer a limited access version
Anthropic also announced that it will open a more broadly authorized version of the Claude Mythos 5 model, which is based on the same foundation as Fable 5, to a small group of cybersecurity and infrastructure providers. The company stated that some protections have been removed in this version. Within the scope of Project Glasswing, which was announced in May, it was shared that Mythos found thousands of critical vulnerabilities in important software and approximately 6200 high or critical vulnerabilities in more than 1000 projects that play a central role in the management of open source and crypto protocols.
