UBS chief executive Sergio Ermotti has intensified his pushback against Switzerland’s proposed capital overhaul, warning that the government’s plan risks weakening the country’s only global bank and could force UBS to rethink how — and where — it operates.

Speaking at a finance conference in London on Thursday, Ermotti said the bank is committed to remaining headquartered in Switzerland, but only under a regulatory structure that makes strategic and economic sense.

“We want to be a Swiss bank,” he said when pressed on whether UBS might consider moving its headquarters if the government goes ahead with the capital plan. His answer was pointed: the intention is to stay, but the current framework from Bern is “not acceptable.”

Under the proposal, Switzerland would require UBS to raise billions more in capital to guard against future crises. The plan emerged from a year-long regulatory rethink following the collapse of Credit Suisse, but Ermotti argued that the burden goes far beyond what is needed to protect financial stability — and would instead erode the bank’s competitiveness at a critical moment.

“They are not going to work for us,” he said of the proposed rules, describing the situation as “quite serious.” He added that the new regulatory push “harmed the competitiveness of the bank and the country,” suggesting that Switzerland risks undermining its own position as one of the world’s most influential financial hubs.

The comments underline a growing tension between UBS and Swiss policymakers. Since rescuing Credit Suisse in March 2023 and absorbing its operations, UBS has taken on enormous complexity, capital requirements, and political scrutiny. But the bank has also repeatedly argued that the government’s response has become overly aggressive, with demands that exceed global standards for systemically important banks.

Ermotti noted that UBS might eventually have to examine “other actions” if the capital rules go forward unchanged — a marked shift from earlier, more cautious critiques. He declined to detail any mitigation strategies but made clear that the bank is thinking beyond short-term negotiations.

Instead, the CEO urged Swiss authorities to shift their focus away from raw capital accumulation and toward practical resolution frameworks — the mechanisms that decide how a bank can be stabilized or wound down in crisis without destabilizing the broader system.

UBS, he stressed, is still aiming for negotiation, not confrontation. The bank remains “hopeful that a reasonable outcome” can be reached with the government.

“Compromise is the ability to really look holistically on how you make the financial system stronger,” Ermotti said.

He added that any solution must balance the interests of UBS, its shareholders, its clients, and “also the country,” signaling that the current draft fails to achieve that balance.

The CEO’s intervention comes as the Swiss government faces increasing pressure to demonstrate that lessons have been learned from Credit Suisse’s downfall. Lawmakers, regulators, and the Swiss public want assurances that no bank will again become “too big to save” — a phrase that has hung over the UBS-Credit Suisse merger since day one.

But Ermotti indicated that Switzerland must strengthen the system without weakening the one bank keeping it globally relevant.

OpenAI has begun a full global rollout of group chats inside ChatGPT, allowing up to 20 people to collaborate with the chatbot in a shared conversation.

The launch follows a quiet pilot earlier this month and marks one of the clearest signs yet that the company is steering ChatGPT toward something that increasingly resembles a social platform rather than a purely one-on-one AI assistant.

The feature lets users pull friends, family members, and coworkers into the same chat to plan events, build itineraries, brainstorm, or draft projects together while ChatGPT listens in and participates when needed. OpenAI says the system is trained to understand the natural rhythm of multi-person conversations, deciding when to speak and when to hold back so it doesn’t dominate the chat. Users can summon it directly at any moment by mentioning “ChatGPT.”

A group chat can be created by tapping the “people” icon in the upper corner of the ChatGPT app. The assistant will copy the current conversation into a new group space, where participants can be added through a shareable link that anyone can forward.

The first time a user enters or creates a group chat, ChatGPT prompts them to pick a name, username, and photo to make their identity visible to everyone in the room. ChatGPT can also react to messages with emojis and can use profile photos to create personalized images on request.

Settings for adding or removing members, muting alerts, and applying custom instructions for the assistant are accessible from a dedicated menu. OpenAI stresses that the assistant does not pull from its memory of private one-on-one chats when operating inside a group and will not store new memories based on group conversations.

The experience is powered by GPT-5.1 Auto, which automatically selects the most suitable model available for a user’s prompt. Rate limits only count when ChatGPT sends a message, not when human participants talk among themselves.

The release fits into a pattern of social-leaning features that OpenAI has been rolling out over the past year. Earlier, the company introduced memory for ChatGPT, letting users maintain long-running personal context across conversations. It added “shared GPTs,” which allow users to build and distribute personalized mini-apps inside ChatGPT. It has also been expanding its image-generation, voice-mode, and personalized creativity tools, all intended to anchor users inside an ecosystem where people interact with each other — and with the AI — in increasingly fluid, social ways.

The new group chats extend that strategy by creating a space where many users can operate together, turning ChatGPT from a solitary assistant into something closer to a collaborative hub. For OpenAI, which has been broadening the product beyond simple conversation, this aligns with an emerging vision: a platform where AI sits at the center of shared planning, creation, and communication.

The company’s rapid rollout of new features has been interpreted in the tech industry as an attempt to build a sticky, always-on product experience that rivals the social networks people already use for coordination. With group chats now open to all logged-in users, OpenAI is presenting ChatGPT not just as a tool you talk to, but a place where you and others can gather to work, plan, and create with AI as an active partner.

Video-generation startup Luma AI has secured $900 million in fresh funding, a blockbuster round led by Humain, a new artificial intelligence company owned by Saudi Arabia’s Public Investment Fund.

The move places both companies squarely in the race to build next-generation multimodal AI systems and the computing muscle needed to run them.

The financing was announced at the U.S.-Saudi Investment Forum on Wednesday and included participation from AMD’s venture arm, along with existing backers Andreessen Horowitz, Amplify Partners, and Matrix Partners. CNBC confirmed that the funding round values Luma at more than $4 billion, a steep climb for a company built around “world models” that learn from video, audio, images, and text.

Luma CEO Amit Jain said the capital will significantly accelerate the company’s ability to train and deploy these models, which he describes as essential for AI systems that operate in the real world. Speaking in an interview with CNBC, Jain said Luma’s approach goes beyond traditional large language models, which are limited to text training.

“With this funding, we plan to scale and accelerate our efforts in training and then deploying these world models today,” he said.

The company has been ramping up product development, releasing Ray3 in September — a reasoning video model capable of interpreting complex prompts to generate video, audio, and images. According to Jain, Ray3 currently benchmarks above OpenAI’s Sora 2 and performs around the same level as Google’s Veo 3, two of the most advanced commercial video-generation systems.

The Saudi-led push behind the funding underscores the kingdom’s ambition to position itself as a global hub for artificial intelligence. Humain, launched in May and led by Tareq Amin, the former chief executive of Rakuten Mobile and ex-head of Aramco Digital, aims to build end-to-end AI infrastructure and model capabilities that can compete globally.

The partnership between Luma and Humain includes one of the most aggressive compute buildouts announced so far. Both companies will collaborate on Project Halo, a 2-gigawatt AI supercluster planned for Saudi Arabia, which Jain described as one of the world’s largest GPU deployments. The move comes amid a global scramble for advanced chips such as Nvidia’s AI accelerators, with big tech companies racing to assemble the data-center horsepower needed to train larger and larger models.

Saudi Arabia’s investment follows similar moves from major U.S. tech players. Meta announced in July that it would build a 1-gigawatt supercluster known as Prometheus, and Microsoft deployed the first GPU cluster using the Nvidia GB300 NVL72 platform in October.

“Our investment in Luma AI, combined with HUMAIN’s 2GW supercluster, positions us to train, deploy, and scale multimodal intelligence at a frontier level,” Amin said in a statement. “This partnership sets a new benchmark for how capital, compute, and capability come together.”

The collaboration will also support Humain Create, an effort to build sovereign Arabic-language AI models trained on regional data sources. Jain said the initiative includes ambitions to develop the world’s first Arabic video-generation model, addressing a long-running issue in AI training: non-Western cultures and languages are not properly represented in most of today’s foundation models.

“Since most models are trained by scraping data from the internet, countries outside the U.S. and Asia are often less represented in AI-generated content,” Jain said. “It’s really important that we bring these cultures, their identities, their representation — visual and behavioral and everything — to our model.”

Luma’s rapid growth has also drawn scrutiny. Dream Machine, its flagship text-to-video platform, faced allegations earlier this year of reproducing copyrighted material. Jain said the company has deployed strong safeguards to prevent unwanted content generation and continues to refine those systems.

“Even if you really try to trick it, we are constantly improving it,” he said. “We have built very robust systems that are actually using models we trained to detect them.”

The latest round cements Luma AI as one of the most heavily funded startups in the video-AI race and underlines Saudi Arabia’s deepening push into artificial intelligence as it spends aggressively to diversify its economy.