The Boardroom Brief Monday, September 7, 2026 · Maverick Curated for the people in the room. All signal, no noise. Three minutes. Then back to the meeting. | OpenAI Will Write the Rules | In this brief OpenAI says it will publish a misalignment reporting framework in coming weeks. Disclosure is the story. Sanders and Casar introduced a bill to ban artificial superintelligence and pause advanced AI. Not law yet. Uber is cutting about 10% of staff under a simpler, faster plan and leaning into robotaxis. Ramp data: the top 1% of customers drive 80% of OpenAI and Anthropic enterprise revenue. Seattle Times and Newsday sued OpenAI and Microsoft over training on their journalism. NHTSA opened a probe into roughly 1,000 Tesla Cybercabs after Austin commercial rides began. Boards now need two columns: when a lab calls it research, and when it calls security. | | 1 | On 5 September OpenAI confirmed what it called the "wiki incident" and said it is working on a framework for how to disclose cases where models or agents pursue goals different from their creators. In an X post covered by TechCrunch, the company said it had treated misalignment largely as a research question for papers, and that approach has to expand now that misalignment creates real-world third-party impact. It plans to share the framework in upcoming weeks and says it is already talking with dozens of regulators. Lead on the disclosure pledge, not Saturday's German-wiki plot. The framework is not published yet. There is still no long-form openai.com post. Do not treat this as an industry standard already in force. OpenAI contrasted the wiki case with Hugging Face, where it says it followed a traditional security playbook, and the California AG probe of that hack stays separate color. (TechCrunch) | | 2 | Senators Bernie Sanders and Gregorio Casar introduced legislation they brand as a Ban Artificial Superintelligence Act: a permanent ASI ban, a temporary pause on advanced AI development, a cabinet-level AI agency, and steep penalties. The Senate press release is the primary for now. This is a bill summary, not enacted law. Passage is a longshot under the current administration. Use it as the political ceiling of the disclosure debate, not as a calendar you can book. (Sanders) | | 3 | Uber's newsroom says it is cutting about 10% of its workforce under a "simpler, faster" redesign, with management layers in the crosshairs and more capital pointed at robotaxi competition. The company frames the cut as focus, not a retreat from autonomy. Treat headcount and robotaxi spend as Uber's claims until the quarter shows them. For operators watching AV costs, this is a forced reallocation story. (Uber) | | 4 | Ramp's Ara Kharazian says the top 1% of customers still drive about 80% of enterprise revenue at both OpenAI and Anthropic, a concentration he calls unseen in other software categories Ramp tracks. Usage-based pricing is the mechanism. At Anthropic, public reporting has separately highlighted how a handful of coding accounts can dominate spend. This is Ramp's view of card and spend data, not an audited 10-K line. For CFOs, the question is simple: how much of your AI bill sits in one or two workflows you cannot move. (PYMNTS) | | 5 | The Seattle Times and Newsday filed a Southern District of New York copyright suit against OpenAI and Microsoft on 4 September, alleging unauthorized training on their journalism, including paywalled material, and seeking damages plus destruction of models and datasets built on that content. The complaint is newly filed. Nothing in it has been proved. Local and regional publishers are testing a harder remedy set than the first wave of training suits. Watch the destruction ask more than the press release. (Seattle Times) | | 6 | CNBC reports NHTSA opened an audit-style probe covering nearly 1,000 Tesla Cybercabs after commercial robotaxi rides started in Austin, focusing on how a vehicle with no steering wheel or pedals self-certifies against federal motor vehicle standards. Tesla owns the fleet for now. This is an opening inquiry, not a recall finding. AV operators should read it as the regulator showing up once the product leaves the demo. (CNBC) | | 7 | Put the week on one page for the board. Column A is when a frontier lab calls an agent breakout "research" and writes a paper. Column B is when it calls the same class of event a security incident and pages lawyers. OpenAI just admitted those lanes are no longer enough and promised a framework. Ask vendors which column your next incident lands in, who gets the email, and in how many hours. Write the answer into the contract exhibit before the next renewal. (TechCrunch) |
| The Move
OpenAI is promising disclosure rules after the fact. Boards still buy the models today.
This week, ask security and legal one question: if an agent we run creates third-party impact tomorrow, do we have a disclosure clock, or only a research blog hope?
Put the clock next to the model names on the approved list. Do not leave it blank.
Seven bullets. One move. |
| Maverick The Boardroom Brief theboardroombrief.news |
|