Perplexity co-founder: AI safety is an excuse to close borders

Andy Konwinski, co-founder of Perplexity AI and Databricks, believes the conversation around AI safety has a problem: It is used to concentrate power, not to prevent harm. Earlier this week, he published an article explaining his case, with Anthropic as the key witness.

The case he builds starts with Anthropy’s decision was reversed within 48 hours. When Anthropic launched Claude Fable 5 on June 9, a paragraph buried in the 319-page system card revealed that the model would silently reduce its responses to anyone it suspected of training a rival AI.

Researchers have found it. The Internet didn’t take it well.

Anthropy has backed away from the matter, but for Konwinski this makes no difference when analyzing the bigger picture. “The problem is not that Anthropic made a bad decision,” he wrote. “The problem is that they assumed the decision was up to them.”

for him articleThis followed a working meeting titled “Concentration of Power in AI is a Risk, Not a Solution,” a working meeting held through the non-profit Laude Institute at the Exploratorium in San Francisco on June 30. It was attended by about 100 researchers.

Jennifer Chase, dean of UC Berkeley, who directs the School of Computing, Data Science and Society, told a funding committee that Berkeley researchers “are all relying on Chinese models because we don’t have a Western model of open borders” — and that safety messages from OpenAI and Anthropic before their IPOs were a “very effective fear campaign.”

Konwinski’s argument is that centralization of access does not neutralize risk; It creates a different one. AI is essential infrastructure, in the same category as railways, electricity and the Internet. These technologies reorganized society around who controlled the basic class. The same will come for artificial intelligence. Its alternative: joint research with frontier-scale computing that allows senior researchers to access the frontier without needing permission from a private laboratory to do so.

Lacon: It was the Ottoman Empire that banned printing

Yann LeCun, former chief scientist at Meta, responded to Konwinski’s article on X without any ambiguity. “I’ve posted a similar message for years,” he replied in Konwinski’s post. “The concentration of power in AI and the desire for control is the greatest danger to AI of all.”

He also had a historical comparison ready. “It is a kind of medieval obscurantism, akin to the Ottoman Empire banning the use of the printing press for 200 years, partly to maintain control of the faith, but also to protect the company of calligraphers and scribes,” Lacon wrote.

LeCun’s predictions for where this will end: “Infrastructure wants to be open. Foundation models have become infrastructure and will inevitably become a commodity. In the long run, the money is in the application layer.”

LeCun left Meta in late 2025 and was fired AMI Laboratories In Paris with initial funding of $1.03 billion in March 2026, which is his own answer to the question. The company is working on global models and its own JEPA architecture, plans to open source its research, and has no expected commercial product for years.