AI’s New Tool Could Speed Up Crypto Hacks, But Also Help Stop Them

Anthropic’s latest release, Claude Fable 5, promises to accelerate code development while sharpening developers’ thinking. The company rolled out two variants:

• Public Claude Fable 5 – Safeguarded against dangerous requests.
• Restricted Claude Mythos 5 – Designed for security professionals.

What the Restricted Model Can Do

• Detect hidden software bugs that are not yet known.
• Chain vulnerabilities into actionable attacks.
• Route risky prompts to a less capable model, Claude Opus 4.8, in fewer than 5% of cases.

Security teams have logged over a thousand hours of testing with no straightforward bypass, though Anthropic acknowledges that determined adversaries may still attempt exploits.

AI’s Double-Edged Sword

Even with guardrails, AI can rapidly uncover weaknesses:

• A hardware‑wallet CTO noted that while controls add friction, they don’t deter a persistent attacker.
• AI can scan codebases, spot misconfigurations, and craft convincing phishing messages in seconds.

Crypto projects routinely lose significant sums to hacks driven by social engineering or human error, rather than smart‑contract bugs. Recent incidents have drained millions from DeFi protocols by exploiting administrative shortcuts or signing mistakes.

The Threat Amplified by AI

Because AI can read public repositories, compare versions, and summarize audits, it becomes a powerful ally for attackers who:

• Spot subtle operational mistakes humans overlook.
• Target private keys, signing flows, and privileged accounts.

Defensive takeaway: Keep private keys in trusted hardware devices that display the exact approval being signed.

Defensive Use of AI

Some projects are already leveraging AI on the defensive side:

• Mapping codebases.
• Running pre‑deployment contract tests.

One DeFi protocol reported that AI helped identify bugs early and improve code quality.

Looking Ahead

The next major hack may not introduce a brand‑new exploit; it could simply be the old tricks—poisoned packages, deceptive developers, flawed signing flows—executed more swiftly thanks to AI. The crypto ecosystem must adapt by tightening operational security and embracing AI as both a threat detector and an attack facilitator.