Inside Invisible

🎤 Invisible brings the multimodal AI debate to SF Tech Week

Invisible will host its first SF Tech Week panel on October 8 at our San Francisco office, co-hosted with Encord and The AI Collective. The event will feature experts from OpenAI, Microsoft, Apple, Encord, and Invisible’s own Lydia Andresen, Director of AI Research. The panel will tackle the toughest questions: how to measure systems when even the biggest labs can’t agree on benchmarks, what personalization means for enterprises, and how to manage risks in fields where mistakes have real-world consequences. We look forward to sharing the main takeaways in a future edition of our newsletter.

Get in line.

‍DOWN TO BUSINESS

🏦 Citi pilots AI agents for research and client profiling

Citigroup has begun a 5,000-person pilot of new agentic AI capabilities within its platform. The update allows staff to direct AI tools to complete multi-step tasks such as researching clients, building profiles from public and internal data, and translating results with a single prompt. Citi Chief Technology Officer David Griffiths said the pilot will assess impact, adoption, and cost-to-value ratios, with built-in controls to manage compute costs as agents take on more complex workloads.

🏛️ Albania appoints AI bot as minister to tackle corruption

Albania has named an AI bot, “Diella,” as its new minister overseeing public procurement. Prime Minister Edi Rama said Diella—whose name means “sun” in Albanian—will manage and award all government tenders, aiming to eliminate bribery and favoritism in a process long plagued by corruption scandals. Diella first launched this year as a virtual assistant, issuing documents with electronic stamps to cut red tape.

🛒 Amazon brings agentic AI to seller operations

Amazon has upgraded its Seller Assistant with agentic AI. It can now monitor account health, optimize inventory, flag compliance risks, and recommend growth strategies. It also integrates with Creative Studio to generate tailored ad campaigns, cutting production time from weeks to hours. Amazon said the AI can also anticipate seller needs, take action with permission, and deliver measurable improvements such as reduced storage costs and higher ad performance.

📞 Enterprises prioritize cloud, AI, and security in voice strategies

A new global survey from AVOXI and Metrigy highlights how contact center voice is shifting toward cloud-first, AI-ready platforms. More than 80% of enterprises are already using AI in voice, but many still face challenges with call quality, data security, and compliance. The report also shows widespread provider churn: 71% of enterprises plan to change their global voice provider, and nearly all are consolidating multiple vendors to cut complexity.

💸 The $3T AI data centre boom strains Big Tech and investors

Global spending on AI data centers is set to hit $3 trillion by 2029, with Big Tech shouldering less than half of the cost. Amazon, Microsoft, Google, and Meta are pouring hundreds of billions into new facilities, but private equity, banks, and sovereign wealth funds are increasingly stepping in to fill a $1.5 trillion financing gap. That shift has turned data centers into a hot asset class, with deals ranging from securitized loans to long-term build-to-suit leases. Analysts warn, however, that obsolescence, energy constraints, and overcapacity could leave smaller players and leveraged developers most exposed, while hyperscalers can better absorb the risks.

HOT MODELS

📊 AI clears toughest CFA exam in minutes

A new study from NYU Stern and wealth-management platform GoodFin found that advanced AI models can now pass Level III of the chartered financial analyst exam, an achievement that usually requires humans about 1,000 hours of study over several years. Earlier research showed AI could handle Levels I and II but stumbled on the essay-heavy final stage. Researchers say the breakthrough signals how quickly AI is advancing in specialized, high-stakes analytical reasoning, though human judgment and context remain critical in financial decision-making.

📂 Claude adds file creation for enterprise workflows

Anthropic has introduced file creation and editing in Claude, allowing users to generate Excel spreadsheets, Word documents, PowerPoint decks, and PDFs directly from prompts. It also allows users to upload raw data and receive cleaned datasets with analysis and charts, turn reports into presentations, or convert invoices into organized spreadsheets with formulas.

🌍 Meta expands Llama AI access to governments in Europe and Asia

Meta is opening its Llama AI models to more governments, including France, Germany, Italy, Japan, South Korea, and organizations linked to the EU and NATO. Governments will be able to fine-tune Llama with their own sensitive data, host models in secure environments, and deploy tailored applications in the field. Meta says the open-source nature of Llama makes it attractive for national security use cases.

💰 Nvidia invests $100B in OpenAI to expand AI capacity

Nvidia will invest up to $100 billion in OpenAI while also supplying the chips needed to run its AI systems. With deliveries set to begin in 2026, the partnership will help overcome the company’s biggest limitation—access to computing power—enabling new breakthroughs and wider use of AI across businesses.

PLOT TWIST

📦“Cleo” lets shoppers ask packaging questions, and get instant answers

At Labelexpo 2025, Digital Link unveiled Cleo, a product assistant built into packaging and activated with a simple QR code scan. Shoppers can ask Cleo questions like whether a cream is safe for pregnancy, what allergens are present, or how to mix a gin cocktail, and get instant answers. For brands, it opens a new direct channel with customers, reducing support costs, providing real-time consumer insights, and turning packaging into an active voice for the brand.

📱 Samsung tests AI shopping on your lock screen

Samsung is partnering with Glance AI to bring a generative AI shopping app to Galaxy phones in the US. The platform can take a photo of a user, generate outfit ideas, and let them purchase clothes directly from the lock screen. The feature is fully opt-in, and being rolled out in phases.

🤖 Viral video shows humanoid robot with lifelike expressions

A video from a Chinese robotics company has gone viral, showing a humanoid robot head that blinks, scans its surroundings, and mimics subtle facial expressions like furrowed brows. The company says its technology aims to build robot heads that can perceive environments and interact naturally with people.

♻️ Subscribe on LinkedIn and share it with your network!