The biggest AI trial ever kicks off
Elon Musk's $130B lawsuit against OpenAI kicked off in federal court, with Musk testifying that OpenAI's for-profit conversion amounts to 'looting a charity.' Meanwhile, Google finalized a classified Pentagon AI deal despite 600+ employee protests, and researchers debuted Talkie, a vintage AI model trained exclusively on pre-1931 text.
Summary
The most high-profile AI legal battle in history began on Tuesday as Elon Musk took the witness stand in his $130B lawsuit against OpenAI and CEO Sam Altman. Musk's suit seeks $130B in damages, the removal of Altman and Greg Brockman from OpenAI's board, and a forced reversal of OpenAI's recent conversion to a for-profit entity. In his testimony, Musk argued that allowing OpenAI's conduct to go unpunished would damage the entire foundation of charitable giving in America. OpenAI's legal team countered that the lawsuit amounts to 'sour grapes,' claiming Musk only objected to the company's structure after it became a successful competitor to his own AI venture, xAI. Microsoft's legal team similarly argued that Musk raised no objections to OpenAI's structure until after its commercial success, and stated it 'knew nothing' of Altman's brief 2023 firing. The trial is expected to last four weeks, with high-profile AI figures set to testify and hundreds of pages of private messages entering the public record.
In a separate but equally contentious development, Google finalized a classified AI deal with the Pentagon the same week that over 600 of its employees published an open letter urging CEO Sundar Pichai to reject military use of its AI systems. The contract, reported by The Information, grants the Pentagon access to Google's AI models for 'any lawful government purpose,' with no legal right for Google to veto specific uses. This follows similar Pentagon deals signed by OpenAI and xAI last month, while Anthropic is currently fighting in court after being blacklisted for refusing to drop its safety guardrails. Notably, Google had scrubbed its no-weapons AI pledge from its principles in 2025, reversing a policy it had adopted in 2018 following earlier employee protests.
On the research front, former Anthropic and OpenAI researchers unveiled Talkie, a 13-billion parameter AI model trained exclusively on approximately 260 billion tokens of pre-1931 text, including books, newspapers, journals, patents, and case law now in the public domain. The project was designed to test how AI reasons when its entire worldview predates the internet and modern computing. A notable finding was that Talkie was able to write working Python code despite Python not existing in 1930, doing so by generalizing from a single example — suggesting deep reasoning capabilities beyond simple memorization. The researchers also noted that Talkie sidesteps a common problem in AI evaluation: benchmark contamination, where models train on their own test data. A GPT-3-level version is reportedly in development.
Other notable news from the edition includes: OpenAI's GPT-5.5, Codex, and Managed Agents becoming available via Amazon Bedrock; NVIDIA releasing the Nemotron 3 Nano Omni multimodal model at 9x the speed of rival open models; the Wall Street Journal reporting that OpenAI fell short of revenue and user growth targets with its CFO questioning spending levels; Anthropic adding new creative app connectors including Blender and Adobe Creative Cloud; and SpAItial launching Echo-2, a text-to-3D world model claiming to outperform World Labs' Marble 1.1.
Key Insights
- Musk argued in federal court that if OpenAI's for-profit conversion is ruled acceptable, it would broadly damage the foundation of charitable giving in America — framing the lawsuit as having societal implications beyond just the tech industry.
- OpenAI and Microsoft's legal teams both argued that Musk's objections to OpenAI's corporate structure only emerged after the company became a direct commercial competitor to his own AI company, xAI, suggesting the lawsuit is retaliatory rather than principled.
- Google's classified Pentagon AI contract grants the U.S. military access to its models for 'any lawful government purpose' with no veto rights for Google — a notable shift given the company scrubbed its no-weapons AI pledge from its principles earlier in 2025.
- Talkie researchers found that the model could write functional Python code despite having no training data from after 1930, demonstrating that the model generalizes reasoning from examples rather than relying purely on memorization of modern programming knowledge.
- The Wall Street Journal reported that OpenAI fell short of its own internal targets for both revenue and user growth, with CFO Sarah Friar raising concerns about the company's massive spending levels — claims OpenAI publicly dismissed as 'ludicrous.'
Topics
Full transcript available for MurmurCast members
Sign Up to Access