AI news, 14 September: SoftBank secures $11.9 billion loan to deepen OpenAI bet

SoftBank locks in upsized loan for OpenAI. SoftBank Group has secured an $11.87 billion two-year loan from about 20 banks to support its investment in OpenAI, exceeding its earlier $10 billion target, according to Bloomberg. The facility forms part of SoftBank’s broader financing push tied to a commitment approaching $65 billion in the ChatGPT maker; the company is also preparing to repay the remaining balance of an earlier $40 billion bridge loan. The deal signals continued bank appetite for AI-linked funding even as industry leaders debate slower development paces and credit risks rise.

Top labs quietly discuss industry safety standards body. Anthropic, OpenAI and Google have held working-group meetings since July on creating an industry-led body for shared testing and pre-release auditing of frontier AI models, The Information reported. Anthropic CEO Dario Amodei is driving the effort; OpenAI’s Sam Altman has backed an independent labs-led approach in the absence of strong federal rules, while differences remain on the degree of government involvement. The talks predate and complement recent public calls for more deliberate pacing of capability gains.

King Charles to host AI leaders in Scotland. Buckingham Palace said King Charles III will convene executives from companies including Nvidia, Google DeepMind, OpenAI and Anthropic, along with officials and ethics advisers, at Dumfries House this week. The gathering aims to explore how AI can be developed and deployed for societal benefit and may consider shared principles, according to the New York Times and other reports. It comes amid heightened public debate over the technology’s risks and rewards.

Microsoft backs deliberate pacing and opens MAI rules to comment. CEO Satya Nadella said Microsoft welcomes the “deliberate pacing” needed to get AI alignment right and will keep superintelligence efforts grounded in human benefit and control. He announced that a Code of Conduct for the company’s first-party MAI models would be published for public consultation, positioning Microsoft alongside other major voices urging stronger safeguards without halting progress.

Anthropic emerges as major customer in large compute deal. The Information identified Anthropic as the previously unnamed buyer in a $13.7 billion six-year GPU/compute agreement with Rum Group, linked to a Georgia data-center site. The arrangement underscores the massive infrastructure demands of frontier AI training even as safety and regulatory conversations intensify.

Court sanctions lawyer over AI-generated fake evidence. The New Mexico Supreme Court fined attorney Stephen Aarons $5,000 and held him in contempt for submitting AI-hallucinated witnesses and fabricated testimony in a murder appeal, The Verge reported. The case highlights growing real-world legal risks when generative AI is used without rigorous verification.