Claude Fable 5 Boosts AI Capabilities, but Also Raises DeFi Security Risks

The new AI model speeds up code analysis and security testing, but it could also make phishing and other attacks on DeFi stronger. Here’s how the industry is dealing with it.

Key Takeaways

• Anthropic launched Claude Fable 5 with stronger reasoning and coding abilities, plus strict safety filters.
• DeFi protocols have already lost more than 840 million dollar this year, mostly because of social engineering and human error.
• AI can help attackers find vulnerabilities faster, but it can also help projects with code analysis and bug detection.

The launch of Anthropic's latest AI model, Claude Fable 5, brings major upgrades in reasoning and coding, but it also creates new security challenges for the crypto market. While DeFi protocols have already lost more than 840 million dollar in hacks this year, the speed and power of this AI tech could raise the risk of attacks.

Powerful AI with Safety Filters

Claude Fable 5 is the first public model in Anthropic's Mythos class and is being praised as its most powerful model yet. The company released two versions: a public version with stricter safety filters and a less restricted version, Claude Mythos 5, which is only available to selected users in cybersecurity and critical infrastructure. These filters are designed to block dangerous use, with risky requests being routed to a less powerful model, Claude Opus 4.8.

Even though extensive testing and outside bug bounty programs did not find a universal way to bypass the system, Anthropic admits that determined and well-funded attackers will keep trying to break through the security. The speed at which AI can identify and exploit weak spots makes it hard to guarantee fully airtight protection.

Risks for DeFi and Social Engineering

The biggest losses in DeFi this year did not come from simple smart contract exploits, but from social engineering and human mistakes. Drift Protocol, for example, lost about 285 million dollar after a long-running campaign that led to admin access, and Kelp DAO was hit by a single-verifier failure that led to nearly 292 million dollar being stolen. Humanity Protocol also suffered a loss of more than 30 million dollar because of a private key compromise.

These examples show that the weak points are mostly in human processes, like key management and signing flows. AI models like Fable can speed up these vulnerabilities by quickly analyzing audit reports, scanning codebases, and generating convincing phishing messages that take advantage of human mistakes.

Ledger CTO Charles Guillemet stressed that AI increases the visibility and speed of these weak spots, but did not create them. He points to the importance of hardware-based security, like private keys generated and stored in certified secure elements with a trusted display for clear signing, to help reduce the risks.

AI as a Double-Edged Sword in Crypto Security

At the same time, some DeFi projects like Pendle are using AI defensively. They use Anthropic's models to analyze their code, run stress tests, and catch bugs early, which helps lead to cleaner and safer smart contracts. Since smart contracts are relatively limited in code and access points, good auditors can still review them thoroughly.

The biggest threat, though, remains the human factor and operational mistakes, with AI mainly speeding up the scouting phase. For European crypto users and developers, this highlights the importance of strong security practices around key management and signing processes, along with using advanced AI tools to spot vulnerabilities early.

These developments show that bringing powerful AI into crypto offers both better security and new risks, especially in the fast-changing DeFi landscape.