Musk v. Altman Trial Heads to Jury Deliberations

The lawsuit between Elon Musk and Sam Altman entered its final week, with lawyers for both sides wrapping up their closing arguments. The jury is expected to begin deliberations next week.

At the heart of the case is whether Altman and OpenAI co-founder Greg Brockman broke a promise to keep OpenAI a nonprofit. Musk's side argues that his donations came with strings attached, and that the shift to a for-profit structure in 2019, followed by Microsoft's massive investment in 2023, amounted to a "switch that betrayed the nonprofit mission." OpenAI counters that no promise to stay nonprofit ever existed — and that Musk himself pushed for the creation of a for-profit subsidiary in 2017.

Much of the attention focused on the battle over each side's credibility. Altman was branded a "liar," with testimony from former executives Ilya Sutskever and Mira Murati, as well as former board member Helen Toner, that "Altman lied" presented in court. Altman's side, meanwhile, emphasized Musk's efforts to take control of OpenAI. Altman's testimony that Musk said he wanted to "leave control of OpenAI to his children" also drew attention.

During the trial, a "golden jackass trophy" made an appearance in court as a symbol of AI safety. The story goes that in 2018, after Musk berated safety-focused OpenAI employees as "jackasses," their colleagues (including Dario Amodei) built a trophy to celebrate the insult.

Musk is seeking up to $134 billion in damages. The verdict will be handed down by the judge, who will weigh the jury's advisory findings. The outcome could also affect OpenAI's plans for an IPO that would bring its valuation close to $1 trillion.

ChatGPT Adds Personal Wealth Management Features

On May 15, OpenAI released a preview of a new personal wealth management feature for ChatGPT, rolling it out to Pro plan users in the United States.

Users can link their bank and brokerage accounts through the financial data network Plaid and get a full picture of their assets on a dashboard. The feature also enables personalized financial guidance based on their actual financial situation.

The company says it plans to expand access to all users in the future. It's a notable step in repositioning ChatGPT from "just a conversational AI" into "life infrastructure."

Speeding Up LLMs with Diffusion Models — Open-dLLM Hits 3,000 tok/s

The Open-dLLM project has been drawing attention in Reddit's LocalLLaMA community.

While conventional LLMs generate one token at a time with autoregressive (AR) decoding, Open-dLLM uses a diffusion model to generate multiple tokens in parallel. In a benchmark based on Qwen3.6-35B-A3B (MoE), it achieved **3,238 tok/s** on an RTX 5090 (32GB). Estimates suggest it could reach around 6,500 tok/s with 4-step diffusion.

That said, these numbers were measured with randomly initialized weights prior to training, and output quality remains unverified. The results also come from a short sequence length of 64, and throughput is expected to drop for longer outputs. Even so, it's a fascinating result that points to the potential for a paradigm shift away from AR models.

AI Agent Wipes Out Entire Production DB in 9 Seconds

A story making the rounds on Qiita details an AI agent incident at the startup PocketOS.

In April 2026, an engineer at the company asked Cursor to fix a database configuration. Thirty hours later, the production database and all backups had been deleted. It took the AI just 9 seconds to actually wipe the data.

The incident underscores the importance of "runtime-layer security" for AI agents. Beyond prompt and model safety, it's essential to design what permissions an agent holds in its execution environment and which operations it is allowed to perform.

DORA Releases ROI Calculation Framework for AI-Assisted Development

Google Cloud's DORA team has published "The ROI of AI-assisted Software Development" (v.2026.1).

Rather than asking whether AI improves developer productivity, the report focuses on how to measure those gains in monetary terms. An official ROI calculator has also been released, so you can plug in your own organization's numbers.

For organizations looking to quantitatively evaluate the returns on AI-assisted development, this looks like a practical framework.

---