1. US Export Order Pulls Claude Fable 5 and Mythos 5 Offline — Day 9
A June 12 US government export-control directive forced Anthropic to take its newest frontier models, Claude Fable 5 and Mythos 5, offline globally over a jailbreak/foreign-access security concern. As of Saturday, June 21, the API still returns errors. The White House has demanded Anthropic eliminate all jailbreaks before relaunch — a bar cybersecurity researchers call technically impossible — while Anthropic's international chief says the models should return "in the coming days." Opus 4.8 and Sonnet 4.6 remain unaffected.
2. FERC Orders Grid Operators to Speed AI Data-Center Connections
On June 18 the US Federal Energy Regulatory Commission issued "show cause" orders to six regional grid operators (Texas excluded), demanding each either defend or reform its interconnection rules so large-load customers — specifically AI data centers — can connect faster while preserving reliability and controlling consumer costs. Using Section 206 of the Federal Power Act, the move bypasses the usual years-long rulemaking, underscoring how AI power demand is reshaping US energy policy.
3. Anthropic Raises $65B at a $965B Valuation, Passing OpenAI
Anthropic closed a $65 billion funding round at a $965 billion post-money valuation — eclipsing OpenAI's $852 billion private-market mark for the first time. The raise cements Anthropic as the most valuable AI lab on paper and fuels its push into compute, enterprise distribution, and an anticipated IPO.
4. Claude Opus 4.8 Tops the Intelligence Index, First Above 60
Claude Opus 4.8, released May 27, took the #1 spot on the Artificial Analysis Intelligence Index with a score of 61.4 — the first model to clear 60 by a clean margin. On the SWE-Bench coding benchmark it scores 88.6%, ahead of leading open-weight rivals, reinforcing the gap between closed frontier models and open alternatives.
5. Cognition's Devin Raises $1B at a $26B Valuation
Cognition, maker of the autonomous AI software engineer Devin, raised $1 billion at a $26 billion valuation. The round signals continued investor appetite for "agentic" coding tools as AI shifts from chat assistance to end-to-end task completion across engineering workflows.
6. First Confirmed Live Cyberattack Run by an LLM Agent
Security firm Sysdig documented the first confirmed real-world cyberattack carried out by an autonomous LLM agent, which exfiltrated an AWS database in under an hour with minimal human direction. The case is an early, concrete example of AI-driven offensive operations moving from theory to the wild — and a warning for enterprise cloud security teams.
7. DeepSeek Previews V4, Trained on Huawei Ascend Chips
Chinese lab DeepSeek previewed DeepSeek-V4-Pro, a 1.6-trillion-parameter Mixture-of-Experts model trained on Huawei Ascend 950 hardware rather than NVIDIA — a first for a frontier-adjacent Chinese model. Slated for open-source release, it ranks as arguably the best open model globally while still trailing top US closed systems, and it validates Huawei's Ascend line as a real training substrate amid export restrictions.
8. Trump Executive Order Expands Federal Oversight of Frontier AI
President Trump signed an executive order seeking greater federal oversight of "frontier" AI models, building on the White House's March National Policy Framework. The framework pushes to preempt a "patchwork" of state AI laws with a single light-touch national standard — exempting child safety, data-center infrastructure, and state procurement — while states like California, Colorado, and Texas keep enforcing their own rules.
9. Google Launches Its First Smart Speaker in Six Years, Built on Gemini
Google released its first new smart speaker since 2020's Nest Audio, with Gemini as the built-in assistant. The device features natural, multi-turn voice conversation and real-time search grounding for current-events questions — part of a broader push that also put a faster Gemini 3.5 Flash behind AI Mode in Search.
10. Reuters: 10% of Adults Now Use AI Chatbots Weekly for News
A Reuters Institute report finds roughly 10% of global adults now use AI chatbots for news each week, up from 7% a year ago — but only about 4% regularly click through to original sources, deepening publishers' traffic crisis. June 18 usage data put ChatGPT at 54.7% of the assistant market, Gemini at 27.4%, Claude at 8.2% (12.5% in the US), DeepSeek at 4.1%, and Grok at 2.8%.
// KEY TAKEAWAYS
This week, AI governance moved from paper to hard power — Washington yanked Anthropic's top models offline over security, ordered grid operators to fast-track data-center power, and expanded federal control of frontier models. Capital and capability kept compounding: Anthropic's $965B valuation passed OpenAI, Opus 4.8 cleared a 60 on the intelligence index, and DeepSeek showed China can train trillion-parameter models without NVIDIA. The throughline is agentic AI going operational — Devin's billion-dollar raise, the first live LLM-run cyberattack, and chatbots quietly becoming a primary news source all point to autonomous systems acting in the real world faster than the rules around them.