Friday's AI landscape is dominated by seismic moves on multiple fronts: Anthropic is heading for Wall Street with a near-trillion-dollar valuation, the White House has drawn new lines around frontier AI oversight, Microsoft is declaring independence from OpenAI, Congress has dropped its most ambitious AI bill yet, and OpenAI is quietly revolutionizing how ChatGPT remembers you.
Anthropic Files Confidential S-1, Eyeing Historic AI IPO
Anthropic, the maker of the Claude family of AI models, has filed a confidential draft S-1 registration statement with the Securities and Exchange Commission, formally kicking off the process to go public. The filing comes just days after the company closed a staggering $65 billion Series H funding round that pushed its post-money valuation to $965 billion — making it the first AI lab to surpass OpenAI in private market value.
The numbers behind the filing are remarkable. Anthropic's annualized revenue run rate crossed $47 billion in May 2026, a roughly fivefold increase from the prior year, driven largely by the explosive adoption of Claude Code, the company's AI coding assistant. The company has committed $1.25 billion monthly to SpaceX through May 2029 for compute infrastructure, underscoring the enormous capital requirements of operating at the frontier.
With OpenAI also expected to file for an IPO in the coming months following its own $122 billion raise in March at an $852 billion valuation, the stage is set for a historic showdown between the two largest AI labs on the public markets.
Trump Signs Executive Order on Frontier AI Oversight
President Trump signed an executive order on June 2 titled "Promoting Advanced Artificial Intelligence Innovation and Security," establishing a voluntary framework for government review of frontier AI models. The order asks AI developers to provide federal agencies with early access to their most capable models for up to 30 days before public release — down from the 90 days proposed in an earlier draft.
The order directs the NSA to develop a classified benchmarking process to evaluate the cybersecurity capabilities of frontier models and to set the threshold for which models qualify as "covered frontier models." Companies that participate voluntarily will be designated as "trusted partners" and gain the ability to collaborate with the government on security testing.
The executive order represents a notable shift in the administration's AI posture, balancing its pro-innovation stance with new national security considerations. While the framework remains voluntary, it creates meaningful incentives for participation through the trusted partner designation.
Microsoft Declares AI Independence with Seven In-House Models
At Build 2026, Microsoft CEO Satya Nadella unveiled seven proprietary AI models under the MAI (Microsoft AI) brand — all built entirely without OpenAI's involvement. The lineup includes MAI-Thinking-1 (a reasoning model), MAI-Code-1-Flash (code generation), MAI-Image-2.5 (image generation ranked above Gemini in Arena evaluations), MAI-Transcribe-1.5 (state-of-the-art transcription across 43 languages), MAI-Voice-2, and two additional reasoning-specialized models.
The headline model, MAI-Thinking-1, was trained from scratch on commercially licensed data with no distillation from OpenAI or any other third-party model family. Microsoft's AI chief described the company as having been "set free" from OpenAI to pursue its own path toward superintelligence.
For a company that spent billions backing OpenAI, the pivot toward self-sufficiency is striking. By running MAI models on its own Azure infrastructure, Microsoft can avoid paying third-party inference costs — a significant margin advantage as AI workloads continue to scale. The models are also available through Fireworks AI, Baseten, and Open Router.
Congress Drops the 'Great American AI Act' — A 269-Page Federal Framework
Representatives Obernolte and Trahan introduced the Great American AI Act, a sweeping 269-page bill that would establish the first comprehensive federal framework for AI governance. The bill's most significant provision is a proposed three-year preemption of state AI laws related to frontier models — a move that would freeze pending state-level regulations, including Colorado's landmark AI Act set to take effect on June 30.
The legislation would require AI developers to publish governance frameworks, report safety incidents, and support a $100 million annual federal standards center. Labor unions have rejected the bill, arguing it prioritizes industry interests over worker protections, while major tech industry groups have praised it as a necessary step toward regulatory clarity.
The timing is notable: Colorado's AI Act, the nation's first comprehensive state-level AI law requiring protections against algorithmic discrimination in employment, healthcare, housing, and other domains, takes effect in just 25 days. The federal bill's preemption clause would effectively pause such state efforts.
ChatGPT Gets a Memory Revolution with Dreaming V3
OpenAI has rolled out its most significant memory upgrade since ChatGPT's launch. The new Dreaming V3 architecture enables ChatGPT to automatically synthesize and catalog user preferences, behavioral patterns, and contextual information without requiring explicit user requests — essentially allowing the model to build a persistent understanding of each user over time.
The update began reaching ChatGPT Plus and Pro users in the United States on June 4, with plans to extend to Free and Go tier users in the coming weeks. OpenAI says the new architecture reduces compute requirements by 5x compared to previous memory systems, enabling the feature to scale across its user base.
However, privacy advocates have raised concerns about the system's approach to behavioral profiling. The automatic nature of the memory synthesis — cataloging preferences without explicit opt-in for each data point — has drawn scrutiny under EU AI Act rules taking effect in August, which impose strict requirements on AI systems that profile users.
In Brief
NVIDIA RTX Spark: Jensen Huang announced an Arm-based superchip at Computex 2026 designed for consumer laptops, with Adobe rebuilding Photoshop and Premiere Pro natively for the architecture. Consumer devices expected autumn 2026.
Alibaba Qwen 3.7 Max: The latest model from Alibaba matches Claude Opus 4.7 on agentic benchmarks at roughly half the input cost and a quarter of the output cost, intensifying the pricing pressure on frontier model providers.
Anthropic Project Glasswing: Anthropic expanded its critical infrastructure security initiative to 150+ organizations across 15 countries, with Claude Mythos Preview identifying over 23,000 vulnerabilities across 1,000+ open-source projects with 90.6% confirmed accuracy.