1. OpenAI wrote its own federal rulebook for frontier AI, then published four documents in a day to sell it OpenAI released four policy documents at once, and read together they describe a company drafting the terms of its own regulation.

2. Amazon will show you clothes that don't exist; law professors picked AI answers 75% of the time Amazon's updated search bar now generates images of products as you type.

3. Google's Gemma 4 12B runs agentic multimodal AI on a laptop, no encoder required The pitch from Google is that you no longer need a server rack to run a capable multimodal model. Gemma 4 12B, released this week under an Apache 2.

In Brief

• Alphabet raised $85 billion in a stock sale to fund Google's AI business Alphabet sold $85 billion in stock, a record raise, to bankroll Google's AI infrastructure and model work. The sale gauges investor appetite for AI-related offerings at scale.

• Microsoft used Build 2026 to push in-house models against OpenAI Microsoft announced in-house reasoning models, a super app, a cybersecurity tool, and AI agents at its Build conference. The lineup positions Microsoft to compete directly with former partner OpenAI. The keynote also covered new Surface hardware and an always-on assistant.

• Microsoft launched Scout, an always-on assistant built on OpenClaw Microsoft Scout integrates into Microsoft 365 apps including Outlook, OneDrive, and Teams. Businesses can assign each employee a virtual assistant for calendars, expense reports, and email drafts. Scout runs across apps rather than living inside Copilot.

• UK regulator ordered Google to let publishers opt out of AI Overviews The Competition and Markets Authority imposed a conduct rule requiring Google to let website owners keep their content out of AI Search features. Publishers can now block their pages from AI Overviews while staying in regular search results.

• Anthropic mapped a year of AI-enabled cyber threats onto MITRE ATT&CK Anthropic published its analysis of how attackers used AI tools over the past year, mapped to the MITRE ATT&CK framework. The report documents observed attacker techniques rather than projected risks.

• Coralogix raised $200 million to monitor AI agents in production Coralogix closed a $200 million round to build observability for AI systems running in production. The company bets demand will rise for tools that track agent behavior, troubleshoot failures, and supply operational data.

• Lovable signed a multi-year deal to expand 5x on Google Cloud Lovable agreed to a 5x expansion of its Google Cloud footprint and wider access to Anthropic's Claude, a source said. The deal spans multiple years.

• Meta opened its WhatsApp Business AI agent to all markets Meta made its AI agent for WhatsApp Business available globally. Meta charges businesses based on token usage rather than a flat fee.

• Google pledged five water commitments amid data center backlash Google laid out five commitments on water use as US communities push back on AI data center construction. The company set a goal to replenish more water than its facilities consume in local areas.

• Flux.ai sent Adafruit a legal demand letter to halt a security report Flux.ai's counsel, Fenwick & West, demanded Adafruit stop publishing an article about Flux, asserting claims under the Computer Fraud and Abuse Act. Adafruit says it accessed only data exposed by a Flux server misconfiguration and has paused its blog while weighing a response.

• Google's Phone app will flag calls from numbers impersonating contacts Google added a feature to Phone by Google that warns users when a caller spoofs a saved contact's number. The app marks such calls as suspicious so users can decline them. The feature targets AI-driven impersonation scams.

• Wasmer built a Node.js edge runtime using OpenAI's Codex Wasmer used Codex with GPT-5.5 to build a Node.js runtime for edge deployment. The company reported shipping in weeks instead of months.

Read the full edition →