Microsoft Says Copilot Is 'For Entertainment Purposes Only'
In what may be the most awkward legal disclaimer in tech history, Microsoft's terms of service for Copilot include a bold-face warning that reads: "Copilot is for entertainment purposes only. It can make mistakes, and it may not work as intended. Don't rely on Copilot for important advice."
The clause, buried in the Copilot for Individuals Terms of Use, drew renewed attention this weekend after journalists and developers flagged the disconnect between this disclaimer and Microsoft's aggressive push to embed Copilot into every corner of its product lineup — from Windows and Office to Azure and GitHub.
Neither Google nor OpenAI apply similar language to their consumer AI products. A Microsoft spokesperson told reporters the company will update the "legacy language," saying it is "no longer reflective of how Copilot is used today." But the incident raises broader questions about how seriously tech companies take the limitations of their own AI tools — especially when billions in enterprise licensing revenue are at stake.
OpenAI Leadership Tensions: Altman vs. CFO Over IPO Timing
Reports surfaced today that OpenAI CEO Sam Altman has sidelined CFO Sarah Friar from key financial meetings, with Friar now reporting to President Fidji Simo rather than directly to Altman — an unusual structure for a chief financial officer at any major company.
At the heart of the friction is OpenAI's IPO timeline. Altman has pushed for a public listing as early as Q4 2026, while Friar has expressed internal concerns that the company may not be ready, citing projected cash burn that could exceed $200 billion before the company reaches positive cash flow. Friar also flagged potential conflicts of interest in the company's recent $122 billion funding round, noting that major investors Amazon and NVIDIA are simultaneously key suppliers.
OpenAI is currently generating over $2 billion per month in revenue. Both Altman and Friar have publicly stated they remain aligned on the company's broader compute strategy, but the organizational reshuffling tells a more complicated story.
Google Gemma 4 Goes Fully Open-Source Under Apache 2.0
Google released Gemma 4 this past week under the Apache 2.0 license — a significant shift from previous Gemma versions, which were open-weight but carried restrictive usage terms. This is the first time Google has released a Gemma model under a truly permissive open-source license.
The Gemma 4 family comes in four sizes — 2B, 4B, 26B, and 31B parameters — with the flagship 31B Dense model ranking #3 worldwide on the Arena AI text leaderboard among open models. Features include context windows up to 256K tokens, native vision and audio processing, and fluency in over 140 languages.
The Apache 2.0 licensing is the real headline here. It allows unrestricted commercial use, modification, and redistribution, putting Gemma 4 on equal legal footing with Meta's Llama models and making it far more attractive for startups and enterprises building production AI systems.
California Charts Its Own AI Regulation Path, Defying Federal Preemption
Governor Gavin Newsom signed a first-of-its-kind executive order establishing new AI procurement standards for state contracts, directly challenging the Trump administration's push for a national AI framework that would preempt state-level regulation.
Under the order, AI companies seeking state contracts must disclose safeguards against misuse — including protections against child sexual abuse material and unlawful discrimination. State officials will be required to watermark AI-generated content, and California will independently review federal supply-chain risk designations rather than automatically deferring to Pentagon assessments.
The move is particularly pointed given the Department of Defense's recent classification of Anthropic as a supply-chain risk. Newsom has made clear that California reserves the right to continue purchasing AI products regardless of federal designations.
Quick Hits
Anthropic's Claude Code leak reverberates: The accidental publication of 512,000 lines of Claude Code source code to npm — caused by a Bun bundler bug and a packaging error — continues to generate discussion a week later. The codebase has been mirrored and forked thousands of times on GitHub, though Anthropic confirmed no customer data or model weights were exposed.
Georgia sends three AI bills to governor: As the state legislature adjourns today, Governor Brian Kemp has three AI-related bills on his desk: a chatbot disclosure and child safety bill (SB 540), an AI study committee bill (SR 789), and a ban on insurance coverage decisions made solely by AI systems (SB 444).
China integrates AI into K-12 education: China is rolling out AI-powered teaching tools across its public school system, with a focus on reducing teacher workload and supporting underfunded rural schools.