AI Safety: Perplexity Co-Founder Accuses Big Labs of Locking Down the Frontier

Andy Konwinski, co-founder of Perplexity AI, is taking direct aim at the giants of artificial intelligence. His target: the “safety” argument wielded by a handful of private laboratories to justify their exclusive control over frontier research.

The Anthropic Fable 5 affair gives him a particularly compelling case study. Behind the rhetoric about existential risks, Konwinski sees above all a strategy of competitive lock-in — and he is far from alone in the tech ecosystem in thinking so.

In a sector where access to the most powerful models directly determines the competitiveness of startups, decentralized protocols, and crypto projects integrating AI, this debate extends well beyond academic circles.

The Fable 5 Affair: When Safety Becomes a Market Argument

Anthropic recently refused to release Fable 5, its frontier model, citing safety concerns. For Andy Konwinski, this decision perfectly illustrates the structural problem facing the industry: a small group of private laboratories is claiming the right to decide who can access the most advanced technologies, in the name of a precautionary principle whose boundaries they define themselves.

Konwinski does not deny that real risks exist around frontier models. But he argues that the safety narrative is being weaponized to erect artificial barriers to entry. In practice, the labs that control the most powerful models — OpenAI, Anthropic, Google DeepMind — benefit directly from a market where their competitors cannot access the same computational resources or the same architectures.

This mechanism is reminiscent, to a significant degree, of the dynamics observed in traditional finance: established players use regulation as a competitive moat, while challengers find themselves blocked by rules designed to exclude them rather than to protect end users.

Open Source vs. Closed Labs: A Strategic Issue for the Crypto-AI Ecosystem

The debate raised by Konwinski resonates directly within the crypto ecosystem. Protocols such as Bittensor (TAO) and decentralized AI projects are built precisely on the assumption that open access to frontier models is a prerequisite for constructing credible alternatives to centralized giants. If closed labs succeed in imposing their vision of “responsible safety” as an industry standard, the room to maneuver for decentralized projects shrinks considerably.

The rise of Meta with LLaMA, and of Mistral AI in Europe, demonstrates that open source can produce competitive models without necessarily sacrificing safety. Konwinski draws on these examples to show that closure is not a technical necessity but a strategic choice — one that primarily benefits those already in a dominant position.

For investors and developers active in the crypto-AI space, the question therefore becomes: in a market where the most powerful models remain locked behind proprietary APIs, which decentralized projects genuinely have the resources to compete? The answer to that question could redefine capital flows toward AI tokens over the coming quarters.

AI Governance: Who Decides, and on What Basis?

At its core, Konwinski’s critique points to a deficit of democratic governance within the AI industry. No transparent mechanism currently defines who can access frontier models, nor the criteria on which those decisions are made. Private labs operate as self-appointed arbiters of a shared technological commons — without public accountability and without any adversarial process.

This situation creates an asymmetry of information and power that the crypto community knows all too well: it is precisely against this kind of opaque centralization that blockchain was designed. The convergence between the decentralization principles championed by Web3 and the demands for open access to frontier AI may well represent the next major ideological battleground in the technology industry.

Perplexity, as an AI search engine challenger to Google and OpenAI, has a direct stake in this debate. But the coherence of Konwinski’s argument — and the Fable 5 precedent — gives it a reach that extends well beyond the immediate commercial interests of his company.