In brief
- Andy Konwinski, cofounder of Databricks and Perplexity AI, warned this week that consolidating AI power poses its own safety risk.
- His essay came after the Open Frontier gathering, a June 30 workshop of about 100 researchers in San Francisco.
- Turing Award laureate Yann LeCun responded on X, likening today’s closed‑lab AI environment to “medieval obscurantism comparable to the Ottoman empire’s 200‑year ban on the printing press.”
Perplexity AI and Databricks co‑founder Andy Konwinski argues that the current AI safety debate is being used to consolidate power rather than mitigate harm. Earlier this week he published an essay making that case, citing Anthropic as a prime example.
His argument hinges on a decision Anthropic reversed within 48 hours. When Anthropic released Claude Fable 5 on June 9, a paragraph buried in its 319‑page system card revealed that the model would quietly degrade responses for users it suspected of training a rival AI.
Researchers uncovered the practice, and the online community reacted strongly.
Anthropic later withdrew the clause, but Konwinski maintains that the episode illustrates a broader issue. “The fault isn’t that Anthropic made a poor choice,” he wrote. “The fault is that they assumed they had the authority to make it.”
His essay titled “Concentration of Power in AI Is a Risk, Not a Solution” followed the Open Frontier workshop he organized through his nonprofit Laude Institute at San Francisco’s Exploratorium on June 30. Roughly 100 researchers attended.
UC Berkeley dean Jennifer Chayes, who heads the College of Computing, Data Science, and Society, told a funding panel that Berkeley scholars are “building primarily on Chinese models because we lack a Western open‑frontier alternative,” and that the safety narratives from OpenAI and Anthropic ahead of their IPOs function as a “highly effective fear campaign.”
Konwinski contends that centralizing access does not eliminate risk; it merely shifts it. AI constitutes foundational infrastructure—akin to railroads, electricity, and the internet—where control of the base layer determines who shapes society. The same dynamic is emerging for AI. His remedy: a shared research commons offering frontier‑scale compute that enables top researchers to reach the cutting edge without needing permission from a private lab.
LeCun: It’s the Ottoman empire banning the printing press
Yann LeCun, Meta’s former chief scientist, replied to Konwinski’s essay on X without reservation. “I have been sharing a similar view for years,” he wrote on Konwinski’s post. “The concentration of power in AI and the drive for control represent the greatest danger posed by AI.”
He also offered a historical analogy. “It resembles medieval obscurantism, akin to the Ottoman empire’s 200‑year prohibition on the printing press, intended partly to preserve doctrinal control and partly to shield the guild of scribes and calligraphers,” LeCun noted.
LeCun’s outlook: “Infrastructure naturally gravitates toward openness. Foundation models are evolving into core infrastructure and will inevitably become commoditized. Ultimately, value will reside in the application layer.”
LeCun departed Meta in late 2025 and founded AMI Labs in Paris in March 2026 with $1.03 billion of seed funding—his response to this challenge. The lab focuses on world models and his JEPA architecture, intends to release its research openly, and does not anticipate a commercial product for several years.
Source link
