AI Industry Daily News

Trump Says US May Take Stakes in OpenAI, Anthropic and Other AI Firms

Speaking aboard Air Force One on June 5, 2026, Trump said the administration is considering acquiring stakes or sharing profits in major AI firms including OpenAI and Anthropic, and plans to convene AI executives at the White House. He framed the idea as letting 'the American people become partners' through small-scale stakes, though no specific percentages or amounts were disclosed.

The move would extend a pattern that began with the administration's roughly $8.9 billion investment for a 9.9% stake in Intel in August 2025, followed by holdings in MP Materials, a golden share in U.S. Steel and stakes in more than ten companies, with committed funds exceeding $10 billion by November 2025.

A meeting with AI executives was reported to be scheduled for early June. The proposal has spread rapidly as an example of the government taking ownership positions in private companies.

Anthropic Reported to Have Filed for IPO

Reports indicate Anthropic has filed for an IPO, a development that could reshape funding dynamics across the AI sector. The company simultaneously released a science blog analyzing why AI has progressed faster in coding than in biology, and has been amplifying discussion of recursive self-improvement risks.

Anthropic's momentum on the product side is also notable: its Claude Opus 4.7 model topped the Android Arena leaderboard, and Claude Code marked one year since general availability while expanding its surrounding infrastructure, including a new observability dashboard for connector developers.

Anthropic Says Biology Databases Are the Bottleneck for AI Agents

In 'Paving the way for agents in biology,' published June 8, 2026, author Laura Luebbert and colleagues compared biological databases to a city designed before cars existed. Non-standard formats, scattered databases, click-based web UIs and ambiguous metadata make them extremely difficult for autonomous agents to navigate, whereas software development tooling was always built to be operated programmatically.

To test the hypothesis, the team built VirBench, a benchmark of 120 queries across 40 pathogens involving complex conditional sequence-retrieval tasks. Frontier models including Claude Sonnet 4, Opus 4.7, Biomni, GPT-5.2-pro and GPT-5.5 struggled with consistency, with accuracy ranging from 16.9% to 91.3% and the same query returning wildly varying result counts across repeats.

Adding a dedicated retrieval tool (gget virus) pushed accuracy to nearly 100% and stabilized reproducibility, leading Anthropic to call for new infrastructure that makes data retrieval reliable for agents.

Google Releases Gemma 4 Open-Source Foundation Model

Google announced Gemma 4 as a new step in its open-source model lineup, emphasizing ease of building high-speed experiences. The release adds to a competitive week for foundation models, with Anthropic's Claude Opus 4.7 leading the Android Arena leaderboard.

The launch comes alongside other Google developer moves, including Chrome 149 DevTools features that let AI coding agents directly operate and verify changes in the browser via the Model Context Protocol.