OpenAI's leaked financials test the AI capex thesis, EU labeling rules arrive in six weeks, and a Copilot flaw exposed 2FA codes across enterprise inboxes.
Leaked audited financials show OpenAI spent $34 billion last year against $13 billion in revenue, producing a $38.5 billion gap. The company has filed confidentially for a public listing that bankers believe could value it above $1 trillion. The filing follows weeks of state AG subpoenas already targeting consumer protection issues at the company.
From August 2, 2026, any business operating in the EU that uses AI to interact with customers or publish content must label that content as AI-generated. Fines reach 15 million euros or 3 percent of global revenue for non-compliance. The rule covers chatbots, marketing copy, product descriptions, and any synthetic media touching EU users.
A patched security flaw in Microsoft 365 Copilot showed how a single email from a stranger could silently pull sensitive data including two-factor authentication codes from a user's inbox without any click or action. The same week, HPE and NVIDIA expanded their joint on-premises AI system to support autonomous agents inside companies, adding hardware, security controls, and governance tools at a moment when most enterprises are deploying agents faster than they can manage them. The two stories define the current enterprise agent risk perimeter.
An independent benchmark of 30 major insurers finds AI has moved from back-office automation into the core of risk decisions, with three firms publicly reporting over $1 billion in combined AI-driven value and the gap between leaders and laggards widening fast. Databricks simultaneously launched CustomerLake, a customer marketing platform built inside its data infrastructure so companies can run AI campaigns without copying sensitive customer data elsewhere. The company also opened a marketplace where enterprise customers install third-party data and AI apps directly inside their own secure environment.
ChatGPT has fallen from over 80% of AI assistant users two years ago to below 50% today, as Google Gemini and Anthropic's Claude take share and monetization becomes as critical as raw user counts. A survey of 2,000 people finds 60% of US consumers are put off when brands use AI as a selling point, while search click-throughs continue to decline. SpaceX has also agreed to acquire AI coding tool Cursor for $60 billion in stock as it rebuilds its AI division after a year of setbacks.
Last 90 days
Non-coders now build the software their businesses run on, ransomware holds at a new higher baseline, and Anthropic sends AI engineers into mid-sized firms.
OpenAI tells companies to price AI by outcome, Cloudflare narrows the web AI agents can read from September 15, and Apple's new Siri skips the EU.
Uber spent its 2026 AI budget by April, OpenAI's new agent deleted files it was not told to touch, and blocked data centers are lifting power bills.
The EU forces SAP to loosen its 22% maintenance fees, and a single US order shows how fast a government can switch off an AI model.
ChatGPT's voice can now search and reason mid-conversation, and experienced buyers stop standardizing on one AI vendor as token spending runs 13 times higher than a year ago.