February 14, 2025 — David Sacks, partner at Craft Ventures and a prominent crypto voice, dropped a quiet bomb on the AI stage. He accused Anthropic, the company behind Claude, of orchestrating a 'regulatory capture' campaign aimed at smothering open-source AI. The accusation itself is not new — the term 'regulatory capture' is a tired policy weapon — but the timing and the accuser make it a signal worth decoding for anyone who understands how power concentrates in permissionless systems.
In crypto, we have a name for this: the security theater of gatekeeping. As a security audit partner who has spent years tracing reentrancy bugs and oracle failures, I recognize the playbook. It starts with a plausible narrative of safety — 'we need rules to protect users' — and ends with a moat that only the well-funded can cross. Anthropic, with its multibillion-dollar backing from Microsoft and Google, is the perfect candidate to write the rules that lock out the decentralized rag-tag of open-source AI developers.
Context: The Hype Cycle Meets the Hammer
David Sacks is not a neutral observer. He is a venture capitalist who has invested in open-source AI projects and famously criticized the 'AI safety' hysteria as a smokescreen for monopoly. His accusation landed on a platform where the crypto-native audience already distrusts centralized authority. The context is a market where AI tokens have surged over 400% in the past year, and where the promise of democratized AI — running models on your own hardware, no gatekeepers — is the narrative that sells.
But the deeper context is the regulatory arbitrage war. Just as DeFi protocols in 2020 exploited the lack of framework to offer unregulated lending, open-source AI models today operate in a grey zone. Anthropic, by contrast, has built its entire brand on 'safety' and 'alignment' — a product that is inherently dependent on compliance. If the regulators adopt a strict licensing regime for AI, only companies with deep pockets and legal teams (read: Anthropic, OpenAI, Google) can survive. The rest become illegal.
Core: The Systematic Teardown of Regulatory Capture
Let me be clear: regulatory capture is not a conspiracy theory — it is a documented economic phenomenon. In every industry where the government sets standards, the incumbents write the standards. The Clean Air Act gave us catalytic converters that only big automakers could afford. The Sarbanes-Oxley Act gave us accounting firms that charged $10 million for audits. Now, in AI, the 'safety' frame is the Trojan horse.
Anthropic's Claude has been marketed as the 'safest' model, with constitutional AI and red-teaming galore. But safety is a relative term. The open-source Llama 3 can be fine-tuned to produce hate speech; Claude can be jailbroken with a prompt about role-playing. The difference is not in absolute safety — it is in the cost of compliance. Anthropic has spent millions on alignment research; the open-source community spends volunteer hours. A regulation that requires every model to undergo a third-party alignment audit would instantly kill the open-source ecosystem, because the audit cost per model would exceed the development budget of 99% of projects.
This is precisely the pattern I saw during the MakerDAO oracle crisis in 2020. When the ETH price spiked, Maker's oracles lagged, causing cascading liquidations. The response from regulators? 'Decentralized oracles need auditing standards.' The actual solution was technical — faster feeds, redundancy — but the proposed solution was regulatory. And guess who could afford to implement the new standards? Only Chainlink, with its centralized nodes and corporate backing. The smaller oracles died. Trust is a variable, never a constant.
Now apply that to AI. The open-source ecosystem is the MakerDAO of artificial intelligence — innovative, decentralized, but messy. Anthropic is the Chainlink — well-funded, compliant, and eager to standardize the market in its image. The accusation of regulatory capture is not just about lobbying; it is about the structural advantage that comes from being the first to define what 'safe' means.
Contrarian: What the Bulls Got Right
I will not be a pure cynic here. The bull case for Anthropic is not entirely self-serving. Open-source AI does pose real risks: bad actors can fine-tune models for propaganda, cyberattacks, or bioweapons. Anthropic's safety-first approach has produced models that are genuinely harder to misuse. Their research on interpretability has advanced the field. And the idea that every AI model should be audited for safety is, in principle, a good one — if the audit itself is not a weapon.
The contrarian insight is that regulation is not the enemy of open-source; capture is. The EU AI Act, for example, explicitly exempts open-source models from many requirements, recognizing that the innovation gains outweigh the risks. The danger is not regulation itself, but regulation that is written by the regulated. David Sacks' accusation highlights a real blind spot: the crypto community often romanticizes open-source as inherently good, ignoring that open-source can also be a vector for harm. The bulls got it right that we need safety guardrails — but they are wrong to believe that Anthropic's version of safety is the only one.
Takeaway: The Audit Is the Weapon
As someone who has audited over 50 DeFi protocols, I know that the auditor is as powerful as the code. If the auditor is captured by the protocol, the audit is a rubber stamp. In AI, the same dynamic applies: if the regulator is captured by the biggest player, the regulation becomes a gate. Smart contracts are deterministic; regulators are not. The solution is not to destroy regulation, but to ensure that the regulatory framework is designed by a diverse set of stakeholders — including the open-source community, the cryptographers, and the users.
Silence in the logs screams louder than alerts. The lack of response from Anthropic to Sacks' accusation is itself a data point. The company has not denied the charge; it has let the silence speak. For now, the crypto community should watch the AI regulatory battle as a dry run for what is coming to DeFi. The same playbook — 'we need standards to protect consumers' — will be used to centralize the decentralized. The ledger bleeds where logic fails to bind.
Every timestamp is a potential crime scene. The one for AI regulation is being written now. If you are not at the table, you are on the menu. Read the source.
