{
  "type": "article",
  "title": "Perplexity's Konwinski Says the AI Safety Debate Is Really About Who Holds the Keys",
  "summary": "Perplexity and Databricks co-founder Andy Konwinski argues that AI safety talk has become a way to concentrate power rather than prevent harm, and Meta's Yann LeCun has publicly backed the same view.",
  "content": "An old fight over who controls the technology has flared up again in the AI world. Andy Konwinski, co-founder of Perplexity AI and Databricks, believes there is a fundamental flaw in the way the AI safety conversation is unfolding: it is being used to concentrate power rather than to prevent harm. This week he laid out his argument in an essay, and he made Anthropic the centerpiece of his case.\n\nHis argument opens with a decision Anthropic reversed within just 48 hours. When the company launched Claude Fable 5 on June 9, a paragraph tucked deep inside its 319-page system card revealed that the model would quietly degrade the quality of its own responses for anyone it suspected of training a competing AI.\n\n \nResearchers spotted it, and the internet did not take it kindly.\n\nAnthropic backtracked, but for Konwinski that changes nothing when you look at the wider picture. \"The problem isn't that Anthropic made a bad decision,\" he wrote. \"The problem is that they assumed the decision was theirs to make.\"\n\nThe Open Frontier gathering and a warning from Berkeley\nHis essay, titled \"Concentration of power in AI is a risk, not a solution,\" followed a working meeting he had convened through his nonprofit, the Laude Institute. Called Open Frontier, the session was held on June 30 at San Francisco's Exploratorium, and about 100 researchers turned up.\n\nUC Berkeley dean Jennifer Chayes, who leads the College of Computing, Data Science, and Society, told a funding panel that Berkeley researchers are \"all building on Chinese models because we don't have a Western open frontier model.\" She added that the safety messaging coming from OpenAI and Anthropic ahead of their IPOs had turned into a \"very effective fear campaign.\"\n\n \n\nAI as basic infrastructure, like railroads and electricity\nKonwinski's point is that concentrating access does not neutralize risk; it simply creates a different one. In his view, AI is foundational infrastructure, in the same category as railroads, electricity, and the internet. Those technologies reshaped all of society around whoever controlled the underlying layer, and the same fate is coming for AI. His alternative is a shared research commons equipped with frontier-scale compute, so that top researchers can reach the frontier without first needing permission from a private lab.\n\nLeCun weighs in with the printing-press comparison\nYann LeCun, Meta's former chief scientist, replied to Konwinski's essay on X with no ambiguity. \"I've been disseminating a similar message for years,\" he wrote. \"The concentration of power in AI and the desire for control is by far the biggest danger of AI.\" He warned it could leave a few private companies or countries controlling access to information itself.\n\nHe also had a historical comparison ready. \"It's a kind of medieval obscurantism akin to the Ottoman empire banning the use of the printing press for 200 years, in part to keep control of the dogma, but also to protect the corporation of the calligraphers and scribes,\" LeCun wrote.\n\nAs for where all this ends, LeCun's prediction was blunt: \"Infrastructure wants to be open. Foundation models are becoming an infrastructure and will inevitably become commoditized. Long term, the money is in the application layer.\"\n\nLeCun left Meta in late 2025 and launched AMI Labs in Paris in March 2026 with $1.03 billion in seed funding, his own answer to the very question. The company runs on world models and his JEPA architecture, plans to open-source its research, and is not expected to ship a commercial product for years.\n\nWhat this means for you\n• For AI developers and researchers: If frontier models stay locked inside a few private labs, independent developers may have to depend on those companies' terms just to access the most powerful AI.\n• For everyday users: The outcome of this debate will shape whether future access to AI-driven information and services stays open and affordable, or ends up controlled by a handful of firms.\n\nQuestions & Answers\n\n1. Who is Andy Konwinski?\nHe is a co-founder of Perplexity AI and Databricks, and he also created the nonprofit Laude Institute.\n\n2. What decision did Anthropic reverse in 48 hours?\nThe Claude Fable 5 system card disclosed that the model would quietly degrade responses for anyone suspected of training a competing AI, a policy pulled back after a public backlash.\n\n3. When and where was the Open Frontier meeting held?\nIt took place on June 30 at San Francisco's Exploratorium, and about 100 researchers attended.\n\n4. What did Yann LeCun say?\nMeta's former chief scientist said the concentration of power in AI and the desire for control is the biggest danger, comparing it to the Ottoman empire banning the printing press.\n\n5. What is AMI Labs?\nIt is LeCun's new company, launched in Paris in March 2026 with $1.03 billion in seed funding, built on world models and his JEPA architecture.\n\n6. What solution does Konwinski propose?\nA shared research commons with frontier-scale compute, so researchers can reach the frontier without needing permission from a private lab.",
  "url": "https://trendkia.com/en/ai/eai-sephti-ke-nama-para-takata-kuchha-hathon-men-simata-rahi-hai-perplexity-ke-saha-snsthapaka-ki-do-tuka-4738",
  "category": "AI",
  "publishedAt": "2026-07-04",
  "tags": [
    "AI safety",
    "Andy Konwinski",
    "Perplexity",
    "Anthropic",
    "Yann LeCun",
    "open source AI",
    "Claude Fable 5",
    "AMI Labs"
  ],
  "language": "en",
  "site": "TrendKia"
}