Callista AI Weekly (September 22 - September 28, 2025)
This week, artificial intelligence moved further from hype to hard results, with global enterprises, tech giants, and policymakers all making significant strides. From factory floors to Swiss boardrooms, AI is now a core business driver—and the world is racing to set the rules for its responsible use. Here’s your comprehensive roundup of the week’s most important AI news, breakthroughs, and governance updates—tailored for Swiss business leaders.
New AI Use Cases
AI is no longer just a buzzword—real companies are reporting measurable gains:
Manufacturing Efficiency
A global manufacturer deployed conversational AI on its factory floor, slashing machine setup times by nearly 30%. The result? Annual savings of around $150,000 per facility. This is not just incremental improvement—it’s a bottom-line boost that scales across operations.
Enterprise Upskilling
Accenture, one of the world’s largest consulting firms, began training its 700,000 employees to use AI agents in daily work. This signals a shift: AI is no longer a side project but a core competency for the modern workforce.
Productivity Tools in Action
Notion, the popular workspace app, revealed that over half its users now engage with its AI features. This surge in adoption has directly contributed to a sharp rise in company revenue, demonstrating that AI-powered productivity is not just a value-add—it’s a growth driver.
Retail and Operations
Walmart’s in-house AI assistant, “Sparky,” is evolving from a simple Q&A bot to an agent capable of making independent operational decisions. This leap toward agentic AI means the system can now optimize processes without constant human oversight.
Banking Innovation
Citi is piloting agentic AI workflows that automatically fetch and summarize data for staff, streamlining information access and freeing up employees for higher-value tasks.
Major Vendor Updates (AI Industry News & New Models)
The AI arms race intensified this week, with major players unveiling new partnerships, products, and models:
Nvidia & OpenAI’s $100 Billion Alliance
Nvidia announced an investment of up to $100 billion in OpenAI, supplying advanced chips for OpenAI’s data centers. This partnership aims to supercharge OpenAI’s computing power and cements Nvidia’s role as the backbone of the AI revolution.
OpenAI’s New Offerings
ChatGPT Pulse:
OpenAI launched ChatGPT Pulse, a feature that proactively generates personalized morning briefings for users—no prompt required. Initially available to premium subscribers at $200/month, Pulse positions ChatGPT as a true smart assistant.GDPval:
OpenAI also introduced GDPval, an evaluation suite that measures AI performance on real-world tasks across 44 occupations, moving the industry closer to benchmarking AI on skills that matter for productivity.
Microsoft’s AI Agents in 365 Copilot
Microsoft rolled out new AI “agents” within its 365 Copilot suite for enterprise customers. These agents, such as the Facilitator in Teams and the Project Manager in Planner, work continuously in the background—drafting agendas, tracking decisions, and assigning tasks. By tapping into company data while maintaining IT security and compliance, Microsoft is shifting from a single AI assistant to a team of specialized, embedded workplace agents.
Google’s Gemini 2.5 Flash & Flash-Lite
Google (DeepMind) released a preview of Gemini 2.5 Flash and a lighter Flash-Lite version on its cloud platform. These models deliver major speed and efficiency gains, with Flash-Lite producing 50% fewer tokens in its answers—making it one of the fastest proprietary models available. For businesses using Google’s Vertex AI, this means more responsive services and lower costs.
Meta’s Llama 2 Approved for U.S. Government Use
Meta’s open-source Llama 2 model was officially approved for use by U.S. federal agencies, marking the first time a free open AI model is cleared for government use. This move gives agencies more control over data and costs, and signals growing trust in open models for serious applications.
Cohere & AMD Partnership
Cohere, an AI startup focused on business applications, raised another $100 million and partnered with AMD to optimize its models for AMD’s GPU chips. This provides enterprises with more hardware choices and positions Cohere as an alternative to OpenAI, emphasizing “AI sovereignty”—keeping data and models under customer control.
Alibaba’s Qwen-3Guard
In China, Alibaba’s AI unit introduced Qwen-3Guard, a set of multilingual safety models designed to act as real-time content filters or “guardrails” for AI systems. This is a step toward making AI outputs safer and more suitable for enterprise and public sector adoption.
AI Governance Developments
AI governance took center stage on the global stage this week, with significant implications for businesses:
United Nations Global Dialogue
All 193 UN member states launched a Global Dialogue on AI Governance in New York—the first time every country is formally discussing how to guide AI’s impact. UN Secretary-General António Guterres emphasized that the question is not if AI will affect international peace and security, but how we manage that influence. The forum aims to develop shared rules for AI safety, ethics, and oversight, reflecting the urgency to coordinate AI policies across borders.
Global Call for “AI Red Lines”
More than 200 prominent scientists, Nobel laureates, former heads of state, and AI pioneers signed a Global Call for “AI Red Lines.” The campaign urges governments to agree by 2026 on certain banned AI behaviors—such as systems that impersonate humans, self-replicate, or autonomously launch weapons. The signatories argue that voluntary ethics pledges are insufficient; legally binding rules are needed to prevent irreversible harm. The analogy to nuclear agreements is clear: even rival nations must agree on red lines to keep AI out of the control of nuclear arms.
Regional Developments
Europe’s comprehensive AI Act remains a reference point and is in its final review stages. In the U.S., discussions about new AI laws and standards are ongoing, with agencies signaling tougher enforcement even in the absence of new legislation.
Breakthrough Research
AI research continues to push boundaries, with several breakthroughs this week that could reshape business and society:
AI Immune System for Cybersecurity
Researchers from Google and academia proposed an “AI immune system” for computer networks, using swarms of AI agents to monitor and defend systems like a biological immune system. In tests, this approach contained cyber threats 3.4 times faster than traditional methods, using less than 10% extra computing resources. For businesses, this could mean cyberattacks are isolated and neutralized almost instantly, dramatically improving corporate security.
Gemini Robotics 1.5 - Agentic AI for Robots
Google DeepMind unveiled Gemini Robotics 1.5, a system that allows a single AI “brain” to control different types of robots. By splitting high-level reasoning from low-level motion, the AI can transfer skills between robots—learning a task in simulation and executing it on various real robots without retraining. This has major implications for manufacturing and logistics, suggesting future robots could be more flexible and quickly taught new tasks.
Improving AI Reasoning
MIT researchers showed that giving large language models more structured feedback can dramatically improve their logical reasoning abilities. By having AI models verify their answers against known rules or external calculators, they became much better at tasks like math word problems and multi-step logic puzzles. This points to a future where AI can plan and reason more reliably—critical for coding, financial analysis, and any domain requiring step-by-step accuracy.
Conclusion
This week’s AI news paints a picture of a technology that is both maturing and accelerating. AI is now deeply embedded in business operations, driving efficiency and competitive advantage for those who embrace it. At the same time, the world is grappling with how to manage AI’s power, with leaders convening to set guardrails and ensure innovation doesn’t outpace our ability to control it.
For Swiss businesses, the message is unmistakable: ignoring AI is no longer an option. The technology is already delivering tangible benefits, but adoption must be paired with caution. Ensuring AI models are secure, unbiased, and compliant with emerging regulations is essential. The age of agentic AI—where autonomous systems handle more decisions—is on the horizon. Success will belong to organizations that leverage these technologies to boost performance, while also keeping a firm eye on ethics, strategy, and the human dimension.
Ready to explore how Agentic AI can transform your organization?
Visit us at https://www.callista.ch/agentic-ai to discover how we can guide your journey into this exciting new era of AI-powered productivity.
Sources
Reuters – “Nvidia to invest up to $100 billion in OpenAI, linking two artificial intelligence titans” (Sept 22, 2025)
TechCrunch – “OpenAI launches ChatGPT Pulse to proactively write you morning briefs” (Sept 25, 2025)
MarkTechPost – “OpenAI Introduces GDPval: A New Evaluation Suite that Measures AI on Real-World Economically Valuable Tasks” (Sept 25, 2025)
Redmond Channel Partner – “Microsoft Updates Copilot To Add Context-Sensitive Agents to Teams, SharePoint” (Sept 23, 2025)
MarkTechPost – “Gemini 2.5 Flash-Lite Preview is Now the Fastest Proprietary Model…” (Sept 27, 2025)
OpenSourceForU / Reuters – “Meta’s Llama AI Approved for Use by U.S. Government Under GSA’s OneGov Program” (Sept 22, 2025)
TechCrunch – “Cohere hits $7B valuation a month after its last raise, partners with AMD” (Sept 24, 2025)
MarkTechPost – Qwen-3Guard release (Sept 26, 2025)
UNRIC – “The artificial intelligence we want” (UN AI governance meetings, Sept 25, 2025)
The Verge – “A ‘global call for AI red lines’ sounds the alarm about the lack of international AI policy” (Sept 22, 2025)
MarkTechPost – “AI Agent Immune System for Adaptive Cybersecurity” (Sept 28, 2025)
MarkTechPost – “Gemini Robotics 1.5… Agentic Robots to the Real World” (Sept 28, 2025)
Radical Data Science – AI news brief (Meta & Hugging Face GAIA benchmark, MIT logic study, Sept 26, 2025)
Organisator – “Über 80 Prozent der Schweizer Firmen nutzen KI ohne Plan” (Sept 24, 2025)
