SunBrief#77: ️Musk v. Altman Trial Begins

Welcome to the SunBrief

Today in SunBrief 🌞

• Find your blind spots in every important meeting

• Judge drops Musk fraud claims against OpenAI

• OpenAI Launches GPT-5.5 as Its Most Capable Work Model Yet

• Stock Updates

• Microsoft Turns Copilot Into an Office Agent

• AI Highlights of the Week

• Too Important to Miss

Find your blind spots in every important meeting

On Call’s new Mac desktop app gives live cues during calls and unbiased feedback after. Spot your blind spots, discover patterns, and improve fast—without extra notes or manual input. Build skills you can take to future roles.

Judge drops Musk fraud claims against OpenAI

Jury will decide whether OpenAI defrauded Elon Musk when it moved away from its nonprofit roots

The long-awaited Musk v. Altman trial is set to begin, with a federal jury asked to decide whether OpenAI misled Elon Musk when it evolved from a nonprofit AI lab into a more traditional for-profit structure. The case could expose private communications between major tech leaders and potentially cost OpenAI billions.

Key Points:

• Core dispute: Musk argues OpenAI was founded as a nonprofit to benefit humanity, but later abandoned that mission by restructuring into a for-profit business.

• OpenAI’s defense: OpenAI says the shift was necessary because building frontier AI required massive capital, and it claims Musk himself supported a for-profit path before leaving the board.

• Musk’s donation claim: Musk once claimed he donated around $100M to OpenAI, but court filings now put the number at about $38M.

• Big remedies requested: Musk wants OpenAI restored as a true nonprofit and has asked for possible monetary damages to go to OpenAI’s nonprofit arm rather than himself.

• Massive financial stakes: Musk’s lawyers are reportedly seeking disgorgement of $65.5B to $109.43B from OpenAI, plus additional claims against Microsoft.

• Unlikely to unwind restructuring: Legal experts say courts may be reluctant to reverse OpenAI’s reorganization, especially after California and Delaware officials already negotiated a compromise.

Why It Matters:
This trial could shape how AI companies balance mission, investors, and public-benefit promises. If Musk wins, it may put pressure on OpenAI and make future AI labs more careful about changing their structure.

OpenAI Launches GPT-5.5 as Its Most Capable Work Model Yet

New model improves agentic coding, computer use, research, cybersecurity, and long-running professional workflows

OpenAI introduced GPT-5.5, its smartest and most intuitive model yet for real work on a computer. It can plan, use tools, check its work, and handle complex multi-step tasks with less supervision.

Key Points:

• Stronger agentic coding: GPT-5.5 reaches 82.7% on Terminal-Bench 2.0, improves on GPT-5.4 in long-horizon coding, and uses fewer tokens across major coding evals.

• Better computer-use workflows: The model performs better at operating real software environments, scoring 78.7% on OSWorld-Verified, helping it click, type, navigate tools, test apps, and complete tasks more independently.

• Professional work upgrade: GPT-5.5 improves documents, spreadsheets, research, business analysis, and workflow automation, reaching 84.9% on GDPval for real-world knowledge work tasks.

• Research and science gains: The model shows stronger performance on genetics, bioinformatics, math, and scientific analysis benchmarks, including improvements on GeneBench, BixBench, and FrontierMath.

• More efficient inference: OpenAI says GPT-5.5 matches GPT-5.4’s real-world per-token latency while delivering higher intelligence and using fewer tokens for many Codex tasks.

• Cybersecurity safeguards: GPT-5.5 is treated as a high capability for both cybersecurity and bio/chemical domains, with stricter classifiers, trusted access for verified defenders, and stronger monitoring against misuse.

Why It Matters:
GPT-5.5 moves AI closer to a true work partner. It can use tools, write code, research, analyze data, and carry tasks forward with less hand-holding. It also shows how AI is shifting from chat assistant to agentic computer worker.

Stock Updates

Microsoft Turns Copilot Into an Office Agent

Copilot can now take multi-step, app-native actions directly inside Office documents, worksheets, and decks

Microsoft announced that agentic capabilities in Word, Excel, and PowerPoint are now generally available, making Copilot a more active collaborator that can directly edit, format, restructure, analyze, and build content inside everyday Office apps.

Key Points:

• Copilot can now take action: Instead of only answering questions or suggesting edits, Copilot can perform multi-step tasks directly on the canvas in Word, Excel, and PowerPoint.

• App-native workflows: In Word, it can draft, rewrite, restructure, and polish documents. In Excel, it can analyze data, create formulas, build tables, and generate visuals. In PowerPoint, it can update decks, refine narratives, and respect company templates.

• Better model capability: Microsoft says recent model improvements in reasoning, instruction following, and multi-step editing made this agentic experience reliable enough for general availability.

• User control remains central: Copilot is designed so users can review, accept, reject, or refine changes, keeping humans in control of structure, style, and brand preferences.

• Early usage gains: Microsoft says the new experience increased engagement, retention, and satisfaction, especially in Excel, where weekly tries per user rose 67%, and thumbs-up satisfaction rose 65%.

• Available now: The features are now the default experience for Microsoft 365 Copilot, Microsoft 365 Premium, and also available to Personal and Family plan users.

Why It Matters:
Microsoft is shifting Copilot from chat assistant to active Office worker. It can now directly edit documents, spreadsheets, and presentations, helping users move from idea to finished output faster.

AI Highlights of the Week

• DeepSeek Unveils V4 AI Model

  DeepSeek has unveiled V4 Flash and V4 Pro, calling them its most powerful open-source AI models yet.

  The models bring stronger coding, reasoning, and agentic performance, plus a 1M-token context window for handling long chats, documents, and codebases.

• OpenAI Introduces Workspace Agents
  OpenAI has introduced workspace agents in ChatGPT, letting teams create shared AI agents for complex tasks and long-running workflows.

  Powered by Codex, the agents can work across tools like ChatGPT and Slack, run in the cloud, follow team processes, and ask for approval before sensitive

• X Launches Stand-Alone XChat App
  X has launched XChat, a stand-alone iOS messaging app for private chats, group chats, file sharing, and audio and video calls.

  The app also adds disappearing messages, message editing, screenshot blocking, and claimed end-to-end encryption, though security experts have raised questions about X’s encryption claims.

• White House Warns of AI Theft by Chinese Firms

  The White House claims foreign actors, mainly China-based firms, are running industrial-scale campaigns to steal US AI technology.

  The memo highlights AI distillation attacks, where firms copy models by exploiting systems, and says the US will coordinate with companies, share intelligence, and develop safeguards to counter these threats.

Too Important to Miss

Anthropic Tests Claude as an AI Shop Manager

OpenAI Launches Free ChatGPT for Clinicians

Google to invest up to $40B to Anthropic

China Blocks U.S. Money for AI Firms

Last Week’s Poll Result

• What defines the next phase of frontier AI?

  Stronger safety controls → 50.00%

  Longer autonomous execution → 26.67%

  Better coding agents → 23.33%

• Should you trust Codex to control your computer for dev work?

  Yes, that’s the future → 44.44%

  Maybe, with supervision → 29.63%

  No, too risky → 25.93%

• What does Qwen3.6 really signal?

  Coding agents are getting cheaper fast → 45.83%
  Efficiency is the new frontier → 33.33%
  Open-source AI is catching up → 20.83%

Feedback

Know someone who may be interested?