The blockchain brain drain just got a poster child, and she's heading to Zuckerberg's AI lab.
The Summary
- Dawn Song, founder of crypto project Oasis, is joining Meta Superintelligence Labs as VP of AI Research, along with other Virtue AI founders in what appears to be an acqui-hire focused on AI safety
- Meta's move centralizes AI security expertise at a moment when the industry is racing to build agents that can act autonomously in the real world
- The hire signals where top technical talent is flowing: not toward building decentralized protocols, but toward whoever can pay for the best AI safety research
The Signal
Dawn Song isn't some crypto tourist cashing out. She's a UC Berkeley professor who built Oasis Network, a privacy-focused blockchain platform that actually shipped product. Now she's leaving that project to lead AI research at Meta, bringing the Virtue AI team with her.
The timing tells you everything. We're entering the era where AI agents don't just answer questions, they execute tasks. Book flights. Move money. Make decisions that used to require a human in the loop. Meta is betting that centralizing AI safety expertise now gives them an edge as that shift accelerates.
"AI's growing allure over blockchain" isn't just about hype cycles. It's about where the actual infrastructure problems are.
Crypto promised to decentralize power. AI is about concentrating capability. Song's career arc mirrors the broader talent migration:
- Top researchers who might have built Web3 protocols five years ago
- Now building guardrails for systems that could actually change how work gets done
- At companies with the compute budgets to train foundation models
The Oasis project continues, but let's be honest about what this signals. When your founder leaves to work on AI safety at a Big Tech lab, the market reads that as a vote of no confidence in the blockchain thesis. Not because blockchain is wrong, but because the more urgent problem is making sure AI agents don't go sideways when you give them real authority.
Meta's acquisition strategy here isn't about buying technology. It's about buying judgment. Song and her team have been thinking about adversarial scenarios, security models, and trust architectures in contexts where there's no central authority to appeal to. That's exactly the skillset you need when you're deploying agents that make consequential decisions.
The Implication
Watch where the security researchers go. That's where the real money thinks the real risk is. Right now, that's AI alignment and safety, not smart contract audits. If you're building in crypto, the question isn't whether blockchain has value. It's whether you can attract talent that has better offers from AI labs.
For companies building agent infrastructure, this is your hiring environment. The people who understand adversarial thinking and security-first design are getting courted by Meta, OpenAI, Anthropic. You're competing for the same pool. Make the work matter, or make the equity life-changing. Preferably both.