πŸ“° Key Takeaways

Mysten Labs co-founder and CTO Sam Blackshear has announced he’s leaving to join Anthropic, where he’ll work on defensive security research. Mysten Labs, the team behind the Sui blockchain, was co-founded by Blackshear and four former Meta executives in September 2021; after his departure, the company’s technical vision will be led by co-founder and CEO Evan Cheng. Blackshear says that as AI is increasingly used to find software vulnerabilities, automate phishing attacks, and speed up attacks on blockchain infrastructure, the balance of power between attackers and defenders is shifting β€” and he believes now is a critical moment to get into security. He points to a turning point: when he ported a static analysis tool he’d originally written in OCaml/Java at Facebook into a version for Move, Sui’s smart contract language, he used Claude to do the port and flag potential vulnerabilities β€” and “it just did it.” That moment made him realize this was an entirely new world in terms of speed and capability, one where triage used to be entirely manual and writing the code itself would have taken a huge amount of time. The piece also notes that crypto companies are seeking access to frontier AI models from companies like Anthropic to strengthen their code defenses, but these models are currently only available to a handful of enterprises. Solana Foundation security chief Michael Coates says that while he understands the rationale for restricting access to advanced models, the vetting and approval process should be simplified so that legitimate defenders can also get their hands on the strongest models β€” because attackers will eventually have tools with comparable capabilities anyway.

πŸ’¬ JudyAI Lab’s Take

Anthropic is putting its most advanced model capabilities directly to work on the defensive side of the security arms race. Sam Blackshear, CTO of Sui development team Mysten Labs, is leaving to join Anthropic to work on defensive security research. He shared that he once took a static analysis tool he’d written in OCaml and Java back at Facebook and had Claude port it to Move β€” and it flagged potential vulnerabilities along the way. “It just did it,” he said, and that made him realize the speed and capability of triage and coding had shifted into a completely different tier, dramatically compressing work that used to depend entirely on manual effort.

This points to a trend worth watching for anyone building with AI: AI is strengthening both sides of the security equation simultaneously, not tilting the field toward one side. Crypto companies are already actively seeking access to frontier models to harden their code defenses, but the strongest models are currently limited to a small number of enterprises. Solana Foundation security chief Michael Coates also noted that while restricting access to advanced models makes some sense, the vetting and approval process should be streamlined so legitimate defenders can access equally capable tools β€” since attackers will eventually get comparable capabilities regardless.

If your project handles sensitive code, now’s a good time to evaluate using AI-assisted static analysis and vulnerability scanning proactively, rather than waiting for a problem to force your hand.

πŸ“… Source Info

πŸ”— Further Reading