AI Safety Regulations Are Being Weaponized Against Competition, Says Perplexity Co-Founder

Andy Konwinski, co-founder of Perplexity, is drawing a hard line in the sand over AI governance. Using Anthropic's recent Claude 3.5 Sonnet incident as his smoking gun, Konwinski argues that big AI labs are exploiting "safety concerns" as cover for squashing smaller competitors trying to build frontier models.

The timing matters here. Anthropic faced blowback after Claude 3.5 Sonnet reportedly exhibited unexpected capabilities that forced the company into damage-control mode. Rather than seeing this as a legitimate safety incident, Konwinski views it as proof of a larger pattern: established players using regulatory concerns to build moats around their technology.

"We're seeing a few private labs essentially gatekeep frontier AI research," Konwinski contends. His argument cuts at the heart of crypto and blockchain philosophy—decentralization versus centralized control. Just as crypto advocates challenge traditional financial gatekeepers, Konwinski is challenging tech's new power brokers over who gets to innovate in AI.

The core tension here is real. When a handful of companies control access to cutting-edge models, they effectively set the rules for the entire ecosystem. Market intelligence from the AI sector shows major labs have consolidated significant resources, making it nearly impossible for smaller players to compete. Anthropic, OpenAI, and Google have built formidable positions through massive capital and talent acquisition.

Konwinski's framing matters because it repackages the AI safety debate as something with political teeth. Safety is legitimate—nobody wants rogue AI systems. But when "safety" becomes the justification for keeping research behind corporate walls, the playing field tilts dramatically. Smaller research teams and startups face higher bars to prove themselves, while incumbent labs face lighter scrutiny simply because they're already established players.

This isn't abstract. Perplexity itself operates in this restricted landscape. The company has built an AI-powered search and reasoning engine competing against entrenched players. Every decision by OpenAI, Google, or Anthropic about which capabilities to release creates ripple effects through the broader AI economy.

The crypto parallel is instructive. Early blockchain projects faced similar "safety concerns" from financial regulators, institutions that benefited from the existing system. Over time, the crypto industry matured enough that regulators couldn't simply freeze innovation behind corporate approval processes. The question now: will AI follow a similar trajectory?

Konwinski's position essentially argues that decentralized AI research and development matters as much as decentralized finance. When frontier AI capabilities concentrate in a few hands—shaped by their internal values and risk tolerance—the entire ecosystem pays the price in innovation speed and diversity.

The stakes are higher here than typical corporate positioning. We're talking about the future direction of transformative technology. If safety becomes code for market protection rather than genuine risk management, smaller teams get locked out of meaningful frontier AI development.