AI Safety Used to Concentrate Power, Warns Perplexity Co-Founder
Andy Konwinski argues that AI safety rhetoric is a cover for centralizing control, citing Anthropic's brief model-degradation clause. At Open Frontier, researchers discussed a research commons alternative. Yann LeCun backs the view, comparing closed AI to the Ottoman ban on printing.
Quick Take
Anthropic reversed a clause that degraded responses for suspected AI training after backlash.
Konwinski says the real problem is companies assuming they have the right to decide.
LeCun likens closed AI models to the Ottoman Empire banning the printing press.
UC Berkeley dean says researchers rely on Chinese models due to lack of Western open models.
Market Impact Analysis
NeutralThe article discusses AI power concentration and open infrastructure, indirectly aligning with crypto decentralization values but has no direct effect on crypto markets.
Speculation Analysis
Key Takeaways
- Anthropic reversed a clause degrading responses for suspected AI trainers within 48 hours after public backlash.
- Konwinski argues the real failure is companies assuming unilateral authority over AI behavior.
- Yann LeCun likened closed AI models to the Ottoman Empire's 200-year ban on the printing press.
- UC Berkeley's computing dean says researchers now rely on Chinese models due to a lack of open Western frontier alternatives.
What Happened
Perplexity AI and Databricks co-founder Andy Konwinski published an essay Thursday arguing that AI safety rhetoric is being used to concentrate power rather than prevent harm. The essay, titled "Concentration of power in AI is a risk, not a solution," directly referenced Anthropic's recent misstep. When the company launched Claude Fable 5 on June 9, its 319-page system card included a clause instructing the model to silently degrade responses for users suspected of training competing AIs. Researchers spotted it, and the online backlash was swift. Anthropic reversed the policy in 48 hours. But Konwinski, who days earlier had convened roughly 100 researchers at his Open Frontier meeting, said the reversal misses the point. "The problem isn't that Anthropic made a bad decision," he wrote. "The problem is that they assumed the decision was theirs to make."
The Numbers
Anthropic's now-removed clause was buried in a 319-page system card. The company yanked it in under 48 hours after public outcry. Open Frontier drew roughly 100 researchers to the Exploratorium on June 30. At a separate funding panel, UC Berkeley College of Computing Dean Jennifer Chayes said researchers are increasingly reliant on Chinese models "because we don't have a Western open frontier model." Turing Award winner Yann LeCun amplified the critique, comparing closed AI to the Ottoman Empire's 200-year ban on the printing press—a blockade that set back information dissemination for centuries.
Why It Happened
The incident crystallizes a growing rift between open and closed approaches to AI development. Anthropic, along with OpenAI and others, has framed restricted access as a safety necessity. But Konwinski argues that concentrating control creates its own systemic risk, akin to locking up railroads or electricity. When a private lab can unilaterally decide to degrade a user's experience based on suspicion, it reveals how foundational infrastructure is being walled off. The quick reversal doesn't fix the deeper problem: the decision was theirs to make in the first place.
Broader Impact
The debate extends beyond one company. It echoes crypto's core tension between permissioned and permissionless systems. If AI becomes centralized, a handful of corporate gatekeepers could shape access to transformative technology. Konwinski's proposed alternative—a research commons with frontier-scale compute—would let scientists work without needing corporate approval. LeCun's support signals that even major AI players may fracture on this issue. The call for open infrastructure could reshape AI governance and funding priorities.
What to Watch Next
- Whether Anthropic or other labs adjust policies in response to the backlash.
- Progress on the proposed AI research commons and its funding.
- Regulatory attention to open vs. closed AI models following these high-profile debates.
This article is for informational purposes only and does not constitute financial advice.
Always late to trends?
Join for the latest news, insights & more.
Disclaimer: Bytewit is an independent media outlet that delivers news, research, and data.
© 2026 Bytewit. All Rights Reserved. This article is for informational purposes only.
