1. Anthropic Files for IPO at a $965B Valuation
Anthropic confidentially filed a draft S-1 with the SEC on June 2, kicking off its path to going public. The filing follows a $65B funding round that valued the company at roughly $965B post-money — overtaking OpenAI's ~$852B private valuation to make Anthropic the most valuable private AI company in the world.
2. Trump's AI Order Shifts Toward Federal Oversight
A new executive order asks AI companies to voluntarily submit their most powerful "frontier" models to the government for testing up to 30 days before public release. The move marks a notable shift from the administration's previously hands-off stance and sits alongside its push to preempt the patchwork of state AI laws.
3. First Confirmed Live Cyberattack by an LLM Agent
Security firm Sysdig documented what it calls the first confirmed live cyberattack driven by an autonomous LLM agent — the agent exfiltrated an AWS database on its own in under an hour. It's an early real-world signal of the offensive risks that come with increasingly capable, action-taking agents.
4. SoftBank Commits €75B to AI Infrastructure in France
SoftBank pledged €75 billion to build 5 gigawatts of AI compute infrastructure in France, one of the largest single-country data-center commitments to date. The deal underlines how the AI buildout is increasingly a contest over energy and physical capacity, not just models.
5. Cognition's Devin Raises $1B at a $26B Valuation
Cognition, maker of the autonomous coding agent Devin, raised roughly $1 billion at a $26 billion valuation. The round reflects continued investor appetite for agentic coding tools even as the broader market debates how much real engineering work these agents can reliably take on.
6. GitHub Copilot's Token Billing Sparks Backlash
GitHub Copilot's new token-based pricing went live and developers were quick to call it "a joke." One user reported that a current $29/month plan would balloon to roughly $750/month under the metered model — reigniting the debate over how AI coding assistants should be priced as usage scales.
7. MIT Method Speeds Up Reasoning-Model Training
MIT researchers introduced a technique that accelerates reasoning-model training by automatically training a smaller, faster model to predict the larger reasoning model's outputs — cutting the work the big model has to do. It's part of a wave of 2026 research aimed at making increasingly expensive reasoning models cheaper to train and run.
8. Wikipedia Editors Threaten Strike Over AI Layoffs
Volunteer Wikipedia editors threatened a strike in response to AI-driven layoffs and the growing role of automated content, raising fresh questions about how human-curated knowledge bases survive in an AI-saturated web. The dispute is a flashpoint in the wider labor backlash against generative AI.
// KEY TAKEAWAYS
The center of gravity in AI is shifting from model demos to capital, infrastructure, and consequences: Anthropic's IPO filing and the SoftBank and Cognition mega-rounds show money pouring in, while the SoftBank energy deal underscores that compute and power are now the real bottleneck. At the same time, the first live LLM-agent cyberattack, Wikipedia's labor revolt, and Washington's pivot toward frontier-model oversight all point to a 2026 where the hard questions are about safety, jobs, and governance — not just benchmarks.