6 events
The big labs pre-empt regulation, the race accelerates
On 18 July, Meta released Llama 2 free for commercial use. Microsoft was named as a preferred partner.
On the 21st, seven major generative AI companies signed voluntary safety commitments at the White House. On the 26th, OpenAI, Google, Microsoft and Anthropic founded the Frontier Model Forum. On the 5th, OpenAI had launched its Superalignment team — research into keeping superhuman AI under human control — and pledged 20% of its compute to it.
On the 11th, Anthropic shipped Claude 2 and opened claude.ai to the public. Elon Musk announced the formation of xAI.
This block is written by the editors. It is kept separate from the sourced record below.
OpenAI announced Superalignment, a new alignment research team aimed at steering and controlling AI systems much smarter than humans. Co-led by chief scientist Ilya Sutskever and Jan Leike, the team would dedicate 20% of the compute OpenAI had secured to date toward solving superintelligence alignment within four years. It drew attention as a major commitment to safety research.
Anthropic released Claude 2 and simultaneously opened claude.ai as a public beta in the US and UK, giving anyone a direct way to use Claude for the first time — previously it was reachable only through the API or partner products. The model improved on coding, math, and reasoning, scoring 76.5% on the multiple-choice section of the bar exam (up from 73.0%) and 71.2% on Codex HumanEval in Python (up from 56.0%). It handled up to 100K tokens across input and output, and Anthropic said it was twice as good at giving harmless responses. API pricing stayed level with Claude 1.3.
Elon Musk announced the formation of xAI. The company had been incorporated in Nevada on 9 March; this was the public announcement. Its stated goal was "to understand the true nature of the universe," and on Twitter Musk put it as forming xAI "to understand reality." The eleven researchers he had assembled came from DeepMind, OpenAI, Google Research, Microsoft Research and Tesla. Musk was a co-founder of OpenAI and had left its board in 2018. Three months earlier he had signed the open letter calling for a six-month pause on giant AI experiments, and his moving to the building side of that argument was received with some surprise. Musk has argued that an AI that wants to understand humanity is less likely to destroy it.
Meta released Llama 2. Unlike its predecessor, which was research-only, it was free for both research and commercial use, with weights and code published for the pretrained models and for versions fine-tuned for dialogue. The announcement was made jointly with Microsoft as preferred partner: Llama 2 was available through the Azure AI catalogue and ran locally on Windows, as well as through Amazon Web Services and Hugging Face. Meta red-teamed the models before release and shipped a transparency schematic, a responsible use guide, and an acceptable use policy alongside them. It was the point at which publishing weights and permitting commercial use came together, and it set the reference for what would be called open-weight distribution — the yardstick still used when OpenAI released gpt-oss in August 2025.
The White House announced that seven leading developers of generative AI had made voluntary commitments on safety: Amazon, Anthropic, Google, Inflection, Meta, Microsoft and OpenAI. Grouped under safety, security and trust, the commitments covered internal and external testing of systems before release, investment in cybersecurity to protect model weights, and work on mechanisms such as watermarking so users can tell when content is AI-generated. The commitments carried no legal force, and the administration presented them as a step ahead of regulation. President Biden met executives from the companies at the White House the same day.
Anthropic, Google, Microsoft and OpenAI announced the Frontier Model Forum, an industry body for the safe development of frontier models. It set out four aims: advancing AI safety research, identifying best practices for responsible development and deployment, sharing knowledge with policymakers and academics, and supporting applications to major societal challenges. The Forum defined frontier models as "large-scale machine-learning models that exceed the capabilities currently present in the most advanced existing models, and can perform a wide variety of tasks". The announcement came five days after the voluntary commitments at the White House, and invited other developers of frontier models to join.