Trump's voluntary AI review order + OpenAI's Codex goes white-collar

President Trump signed an executive order on June 2 asking AI developers to voluntarily share frontier models with the federal government up to 30 days before public release for national-security and cybersecurity testing. The order explicitly bars any mandatory licensing or pre-clearance regime, and was cut back from an earlier 90-day window after industry objections, including from former White House AI czar David Sacks. It also stands up a voluntary cyber "clearinghouse" coordinated by the US Treasury Department and directs the Justice Department to prioritize AI-assisted hacking cases.

OpenAI launched six role-specific Codex plug-ins — covering data analytics, creative production, sales, product design, equity investing, and investment banking — each bundling the integrations, instructions, and context to approximate a specific job. A new "Sites" feature turns Codex output into hosted interactive webpages via partners like Figma and Replit. OpenAI says knowledge workers are now roughly 20 percent of Codex users and growing more than three times as fast as its developer base.

At Build 2026, Microsoft introduced Scout, an always-on agent that plugs into Microsoft 365 — Teams, Outlook, calendar, and email — and proactively handles meeting prep, scheduling conflicts, and routine tasks. It is built on the open-source OpenClaw project (whose founder was hired by OpenAI earlier this year) plus Microsoft's WorkIQ. Each Scout runs under its own governed Entra identity rather than a shared service account, and is rolling out in private preview to Frontier organizations.

Microsoft also used Build to release MAI-Code-1-Flash, a small, efficiency-focused coding model trained on its own production GitHub Copilot workflows. The company says it outperforms Anthropic's Claude Haiku 4.5 across its coding benchmarks — including a roughly 16-point edge on SWE-Bench Pro, at 51 percent versus 35 percent — while using up to 60 percent fewer tokens. It is rolling out now to GitHub Copilot users inside Visual Studio Code, with no setup required.

Google began a global rollout of fake-call detection on Android 12 and later, starting with Pixel. When two people both use Phone by Google, the caller's device sends a silent, encrypted verification signal; if a scammer spoofs that contact, the missing signal triggers a warning to hang up. The feature is on by default and targets a fast-growing threat — INTERPOL tied impersonation fraud to over $400 billion in global losses this year.

Following Alphabet's plan to raise $80 billion in equity for AI compute — with Berkshire Hathaway taking $10 billion — Ben Thompson's Stratechery argues the move marks a pivot from a high-margin advertising business into a capital-intensive infrastructure company, akin to how Berkshire redeployed See's Candies profits into railroads. His thesis: in a compute-constrained market, the firm with the most cash to deploy compounds its lead, and Berkshire's stake validates that bet.

A Stanford-led study ran nearly 3,000 blind comparisons of AI-generated and professor-written answers to contract-law tutoring questions, and the AI responses won 75 percent of the head-to-head matchups across 16 law professors. Crucially for classroom use, reviewers flagged AI answers as potentially harmful just 3.5 percent of the time, versus 12 percent for human-written peer answers. The authors frame the results as evidence that well-built systems can widen access to high-quality legal-education tutoring.