The History of AI

The race for AGI

The race for artificial general intelligence (AGI): what the term means, the labs' stated missions, safety and regulation, compute and capital, and whether AGI is already here.

The concept

  • . Mark Gubrud uses the phrase "artificial general intelligence" for the first time
  • . Shane Legg and Ben Goertzel revive the term and make it popular
  • . Google DeepMind researchers propose five levels of AGI, from "emerging" to "superhuman"

The missions

  • . DeepMind is founded with the goal of building general intelligence (DeepMind is founded)
  • . OpenAI is founded. Its mission: "to ensure that artificial general intelligence benefits all of humanity" (OpenAI is announced)
  • . Anthropic is founded, with a focus on the safety of advanced systems (Anthropic is founded)
  • . Elon Musk founds xAI
  • . Ilya Sutskever founds Safe Superintelligence Inc.
  • . Meta creates Meta Superintelligence Labs, led by Alexandr Wang, after investing $14.3 billion in Scale AI

Safety and regulation

  • . The heads of OpenAI, Google DeepMind, and Anthropic sign a statement that mitigating the risk of extinction from AI should be a global priority
  • . Anthropic publishes its Responsible Scaling Policy, which ties safety levels to a model's capabilities
  • . The Bletchley Park summit: 28 countries sign the first international declaration on the risks of advanced AI. Seoul (2024), Paris (2025), and New Delhi (2026) follow
  • . The European Union's AI Act enters into force. Its obligations for general-purpose models apply from August 2025

Compute, capital, and the state

  • . The United States restricts exports of advanced chips to China (Chip export controls on China)
  • . Stargate: up to $500 billion in data centers (Stargate)
  • . DeepSeek-R1 sends Nvidia down 17 percent: a Chinese lab nears the frontier at a far lower reported cost (DeepSeek-R1)
  • . OpenAI becomes a public benefit corporation
  • . SpaceX buys xAI in an all-stock deal that values the combined company at $1.25 trillion
  • . The US Department of Defense demands unrestricted use of Claude. Anthropic refuses, is designated a supply chain risk on March 5, and a federal judge grants an injunction on March 26
  • . OpenAI and Microsoft revise their agreement and drop the clause that hinged on declaring AGI
  • . Anthropic raises $65 billion at a $965 billion valuation. OpenAI had closed a $122 billion round at an $852 billion valuation
  • . Anthropic (June 1) and OpenAI (June 8) file for their IPOs. OpenAI later pushes its offering back to 2027
  • . The US government restricts access to Claude Mythos 5 and Fable 5 on national security grounds; the restriction lasts until June 30 (Claude Fable 5 and Claude Mythos 5)

Has it already arrived?

  • . In a preregistered study by Jones and Bergen, people take GPT-4.5 for a human in 73 percent of five-minute conversations: the authors present it as evidence that it passes a standard Turing test
  • . Sam Altman says OpenAI has built AGI, that the milestone has already gone by, and that it is time to talk about superintelligence
  • . Unveiling GPT-6 Astra, Greg Brockman says it is not far-fetched to feel we are in the era of AGI. At the same event, OpenAI's chief scientist warns that progress in intelligence does not guarantee progress in alignment (GPT-6 Astra)