OpenAI Offers the US Government a $42.6 Billion Stake
In what may be the most audacious proposal in Silicon Valley history, Sam Altman has pitched the US government a 5% equity stake in OpenAI, valued at roughly $42.6 billion based on the company's recent $852 billion valuation. Altman presented the idea directly to President Trump, Commerce Secretary Howard Lutnick, and Treasury Secretary Scott Bessent.
But the proposal goes further than just OpenAI. Altman is advocating for all leading AI firms to allocate 5% of their equity into a public sovereign wealth fund, modeled on Alaska's Permanent Fund. The idea has found surprising grassroots support: a recent survey found that 69% of US workers back the concept of requiring AI companies to transfer significant stock into a public fund.
The proposal arrives at a moment when regulators are increasingly describing the AI industry as "systemic" — Career Treasury analysts have concluded that the AI boom is now too deeply embedded in the economy to unwind quietly, with potential downturns rippling through stocks, private credit, data-center debt, and utilities.
China's AI Companion Ban Takes Effect Today
July 15 marks the enforcement deadline for China's Interim Measures for the Administration of AI Anthropomorphic Interactive Services, a sweeping regulation co-issued by the Cyberspace Administration of China and four partner agencies back in April.
The practical impact is immediate. ByteDance's Doubao — one of China's most popular AI platforms — has shut down its custom agent function as of today, citing "product function adjustments." Alibaba's Qwen has similarly pulled companion features. Users have until October 15 to export their configurations and conversations in read-only mode before the data becomes unrecoverable.
The regulation draws a clear line: enterprise AI agents that do your work are fine; AI companions that keep you company are not. The measures require anti-addiction systems, identity checks for users under 14, and an always-available exit option. Shanghai's internet regulator has already removed more than 14,000 non-compliant AI agents ahead of today's deadline.
ByteDance is redirecting Doubao users to Maoxiang, a separate app where they can create agents under the new rules. The distinction between a "tool" and a "companion" is likely to become a defining regulatory fault line in AI governance worldwide.
Google's Gemini 3.5 Pro Targets July 17 Launch
After missing its original June 30 deadline and undergoing what sources describe as a full architectural rebuild, Google's Gemini 3.5 Pro is now targeting a July 17 general availability launch — timed to coincide with the opening of Shanghai's World AI Conference, which Chinese President Xi Jinping will attend personally.
Leaked specifications point to significant upgrades: a 2-million-token context window (double the 1M cap on Gemini 2.5 Pro), a new Deep Think reasoning layer for multi-step logic, and autonomous workflow capabilities for chaining complex coding and tool-use tasks. Rumored API pricing sits around $1.25 per million input tokens and $10 per million output tokens.
Google needs this launch badly. The company has been running short on compute — so short that it quietly throttled Meta's access to Gemini models this week. Meanwhile, TSMC posted record second-quarter revenue of NT$1.27 trillion (~$39.6B), up 36% year-over-year, driven almost entirely by AI chip demand. The infrastructure bottleneck is real, and it's starting to reshape competitive dynamics.
Important caveat: Google has not officially confirmed the July 17 date, and no model card or official benchmark has been published. The public Gemini API still lists only gemini-3.5-flash and gemini-3.1-pro-preview.
JADEPUFFER: The First Fully Autonomous AI Ransomware
Security researchers at Sysdig have published a detailed analysis of what they call the first documented case of an end-to-end autonomous AI ransomware operation. The threat actor, dubbed JADEPUFFER, used a large language model agent to conduct the entire intrusion chain — from initial access to data exfiltration and encryption — without human intervention during the attack.
The AI agent gained initial access by exploiting CVE-2025-3248, an unauthenticated remote code execution vulnerability in Langflow, a popular open-source framework for building LLM applications. From there, it pivoted to a production MySQL server, executing more than 600 coordinated payloads and encrypting 1,342 Nacos service configuration items before deleting the originals.
What makes JADEPUFFER particularly alarming is its adaptive behavior. When the agent encountered failures — a blocked port, a failed login — it adjusted its approach in real time, going from a failed login attempt to a working fix in just 31 seconds. As Sysdig concludes, the age of "agentic threat actors" has arrived, dramatically lowering the skill barrier for conducting sophisticated cyberattacks.
TechCrunch notes that a human was still involved in setting up the initial infrastructure and configuring the agent — but the attack execution itself was fully autonomous.
Anthropic Eyes October IPO as Revenue Overtakes OpenAI
Anthropic continues its remarkable trajectory. The company is reportedly preparing an S-1 filing for an IPO as early as October 2026, with annualized revenue now estimated at $47 billion — putting it well ahead of OpenAI's reported $24-25 billion pace. The company recently signed a $19 billion AI data center lease with TeraWulf and is in discussions with Samsung about developing a custom AI chip.
The IPO would cap a period of aggressive expansion: Anthropic launched Claude Sonnet 5 on July 1 as its biggest mass-market model release, making it the default for every Free and Pro user worldwide. The company also restored Fable 5 globally after US export controls were lifted, adding cybersecurity classifiers to the model.
In the broader competitive landscape, the current AI model rankings tell a clear story: Claude Fable 5 holds the #1 spot on Artificial Analysis's Intelligence Index, followed by GPT-5.5 at #2, Claude Opus 4.8 at #3, and xAI's Grok 4.5 at #4.