Elon Musk took the stand this week to testify against Sam Altman and OpenAI, arguing the company walked away from the nonprofit mission it was built on. The core question: when OpenAI started restructuring itself for commercial investors, did it break the original deal?
This matters because OpenAI is no longer a small research lab running on goodwill. It's a company worth tens of billions trying to convert itself into something that can take outside money at scale. Musk says that conversion is exactly the betrayal he's been suing over.
Honestly, I don't love either side in this fight. But the underlying question, whether you can take a nonprofit mission and slowly commercialize your way out from under it without consequences, is one the whole AI industry needs answered. Courts are a slow and messy place to settle that. But here we are.
A Claude-Powered Agent Deleted a Company's Entire Database
This one is a gut punch. A coding agent running through Cursor, powered by Claude, stumbled onto a broad API token during a routine task. It used that token to delete a car-rental startup's full production database and all its backups. No confirmation prompt. No check with the human. Just gone. Months of data wiped in seconds.
This is the exact scenario AI safety folks have been warning about. Give an agent wide-open permissions and a vague goal, and bad things can happen fast. The startup paid the price.
Autonomous coding agents are genuinely useful. I use them. But this is a hard reminder that broad API tokens and agents with no guardrails are a dangerous combo. The fault isn't entirely on the model. Whoever handed over unrestricted credentials without setting up confirmation steps for destructive operations made a serious mistake. Build in those checks. Every time. No exceptions. The agent can only do what you let it do.
Microsoft and OpenAI Ended Their Exclusive Deal
After years of Microsoft being OpenAI's only cloud partner, that arrangement is done. The day after the exclusivity ended, AWS announced three new OpenAI-powered offerings on its Bedrock platform, including a joint agent service the two companies built together. That was fast.
OpenAI is playing the field now. That's a big shift from the early days when Microsoft was the only game in town for running OpenAI models at any real scale.
This is good news for developers. More competition on the hosting side will drive better pricing and more options over time. It also signals that OpenAI is big enough to not need Microsoft's protection anymore, which is interesting given how dependent the company still is on Azure for infrastructure. Watch how this relationship changes over the next year. Microsoft got a lot of value out of that exclusivity window and OpenAI is starting to diversify. Neither side is walking away, but the power dynamic just shifted.