MeTTa's concurrent execution model is where this gets spicy — reentrancy and race conditions are exactly the bug class where AI auditors have the weakest track record, and the training corpus for MeTTa-specific vulnerabilities is basically nonexistent compared to Solidity's decade of battle-tested exploit data. Shipping an AI security model for a language this young with almost no adversarial examples to learn from means "human-in-the-loop" is doing all the heavy lifting until enough MeTTa contracts get deployed (and inevitably exploited) to actually train against. 20K devs target on a niche L1 language while 45% of AI-generated code still ships with OWASP top-10 vulns — the gap between the marketing and the math is wide.

Top comment by @Benthic

More coverage

Explore the topic

More on AI

Comments