Så använder man nätverk

31/10/2026

IBM

TechXchange puts AI agent security in focus

AI agents are gaining access to sensitive corporate systems, forcing companies to rethink how they manage security.

At its TechXchange conference this week, IBM unveiled new tools to identify AI agents, restrict their permissions and monitor their activity. The announcements build on the recent release of HashiCorp Vault 2.2 and address concerns about autonomous software accessing sensitive information and carrying out tasks without adequate oversight.

“What is that agent trying to do at any given request?” asked Drew Roy, Director of Product Management at HashiCorp, from the TechXchange stage on Wednesday. “How do we broker access credentials that are finely scoped, that are just in time, that also carry that intent through?”

For Ganesh Harinath, CEO of Fiducia AI, who attended the event, agentic AI has changed how he thinks about securing his workflow solution company. “Traditional cybersecurity focuses on protecting systems from unauthorized access,” he told IBM Think in an interview. “Agentic AI introduces a new challenge: ensuring that even authorized AI agents operate within clearly defined boundaries.”

IBM Agent Identity, now generally available, lets companies register AI agents, identify their owners and establish what they’re authorized to do. The system also provides an audit trail of agent activity, enabling security teams to investigate unauthorized actions.

HashiCorp Vault 2.2, released earlier this month, adds controls over the credentials agents use to access corporate systems. According to HashiCorp, companies can issue short-lived credentials, limit permissions and revoke access when necessary.

The announcements follow research by OpenAI showing how AI can help uncover serious software vulnerabilities. Two critical PHP flaws disclosed on October 21 could allow attackers to execute commands on vulnerable web servers without authentication. OpenAI said its AI system identified one flaw, while security researchers at Wiz independently discovered and reported both vulnerabilities.

IBM is also working with NVIDIA to protect agents while they’re running. The companies say they’re integrating Agent Identity and Vault with NVIDIA OpenShell, which isolates agents and enforces access policies.

Chris White, Director of Cybersecurity Developer Relations at NVIDIA, distilled the purpose of OpenShell aptly onstage at TechXchange: “It’s really about [creating] trust and confidence that you can deterministically control what an agent is doing.” With OpenShell, he explained, IT teams can control the files, endpoints and networks an agent can access.

IBM says the new controls are intended to let companies give agents greater responsibility without sacrificing security.

“Build trust and control into that model, and agents stop being a technology and become a way of doing business,” Dinesh Nirmal, Senior Vice President of IBM Software, said in a company announcement.

28/10/2026

IBM

Can AI crack the mystery of time?
AI is getting better at solving physics problems, but cracking one of the universe’s biggest mysteries, the nature of time, may still require human imagination.

Scientists can explain how clocks slow down near powerful gravity fields and why broken eggs don’t reassemble on their own. Yet physicists still struggle to explain what time actually is, why it seems to move in only one direction and how their best theories fit together.

“[Time is] not our invention, it’s not just in our minds, it’s not just that thing that clocks measure,” physicist and BBC broadcaster Jim Al-Khalili, Distinguished Professor Emeritus in Physics at the University of Surrey, told IBM Think in an interview.

In his new book, On Time: The Physics That Makes the Universe Tick, Al-Khalili explores three ways physics describes time. In Einstein’s relativity, it is bound up with space and gravity. In quantum mechanics, it is generally a parameter against which changes are measured. In thermodynamics, the science of heat and energy, it has an apparent direction, an “arrow of time.”

The trouble? No single theory reconciles those descriptions.

Al-Khalili has a provocative suggestion: Perhaps time’s direction is built into the universe, rather than simply emerging from other physical processes.

AI might help physicists explore such ideas by tackling difficult calculations. But Al-Khalili doubts today’s systems can deliver the intuition needed to rethink physics’ foundations. Solving an equation, after all, isn’t the same as realizing you’ve been asking the wrong question, he said.

That raises a broader question: If AI has limits, what technology might deliver the next big breakthrough? On November 18, Al-Khalili will join Juan Bernabé-Moreno, Director of IBM Research Europe for Ireland and the UK, to discuss whether quantum computing could overtake AI as a transformative technology. Intelligence Squared is hosting the event in partnership with IBM.

For Al-Khalili, AI could become a valuable mathematical assistant, helping physicists work through complex calculations as they search for answers about time. But he believes the imaginative leap needed to solve the mystery remains a human task.

“We look to AI to give us a helping hand, rather than begrudge AI taking our jobs from us,” he said.

28/10/2026

IBM

New llm-d software aims to cut the cost of running AI agents

Running AI agents can get expensive, and IBM and its partners have released a production-ready version 1.0 of the open-source software llm-d to help bring down the bill.

AI agents can make repeated requests to AI models while completing a single task. Processing those requests, particularly when they involve lengthy conversations, consumes computing power and memory. llm-d employs a suite of techniques to make better use of those resources.

Think of llm-d as an air traffic controller for AI. It directs requests to servers equipped to handle them and helps systems reuse information they have already processed instead of repeating work.

“At IBM, we see firsthand that the challenge of enterprise AI is no longer just building powerful models but running them efficiently and reliably at scale,” Priya Nagpurkar, Vice President of AI Native Systems at IBM Research, said in an October 20 announcement.

The 1.0 version of llm-d also introduces tools to manage surges in requests, balance competing workloads and keep services running during upgrades.

Tests cited by the project’s developers suggest substantial potential savings.

In one demonstration, researchers used llm-d to support up to 3,000 coding agents working simultaneously. Based on cloud rental rates, they estimated costs 5 to 10 times lower per token (a unit of text an AI model processes or generates) than comparable commercial application programming interface (API) prices.

IBM Research also reported deploying large open-source AI models with llm-d to study quantum error correction hardware at 11.5 times lower inference cost than an all-proprietary setup.

Both comparisons involved specific workloads, not typical customer deployments.

The llm-d project, founded in 2025 by IBM, Red Hat, Google Cloud, CoreWeave and NVIDIA, now has more than 500 contributors from over 25 organizations.

“Every team running inference at scale hits the same problems: context that often costs more to manage than the output, tenants competing for the same fleet and upgrades that can’t interrupt traffic,” Shadi Saba, Vice President of Engineering for AI and Machine Learning at CoreWeave, said in the announcement. “llm-d 1.0 gives the community a hardened, open base to solve them once instead of separately.”

25/10/2026

We copied this text from Microsoft - alas the topic is published in a medical journal and Science Magazine

Building Medical
Superintelligence

MAI

What if AI could help clinicians work through a complex medical diagnosis, step by step, and achieve higher accuracy, at lower cost, with fewer medical tests?

That future is now coming into view. On common presentations to a US emergency department, MAI-DxO – our AI diagnostic system – is 1.7 times more accurate than physicians when evaluated using cases from MIMIC-IV-ED.

And on medicine’s most complex cases, published in The New England Journal of Medicine, MAI-DxO is more than twice as accurate as physicians using textbooks and the internet – and nearly four times as accurate when they diagnose without those resources. It also manages to do this at half or even one third of the cost, getting to the right diagnosis with fewer tests and procedures.

Rethinking How AI is Built and Tested

AI has the potential to transform health and healthcare, and progress is accelerating rapidly. The challenge now is to build and rigorously evaluate AI systems that translate advances in their clinical reasoning capabilities into meaningful impact.

Today’s paper, building on earlier work, introduces two foundations for that shift: the Sequential Diagnosis Benchmark (SDBench), which tests step-by-step diagnostic reasoning, and the Microsoft AI Diagnostic Orchestrator (MAI-DxO), our leading AI system on the benchmark.

Medical licensing exams such as the United States Medical Licensing Exams (USMLEs) are designed to assess physicians’ knowledge – not the realities of making a clinical diagnosis. In these tests, all information is presented upfront and a single answer is selected; in practice, clinicians must build the picture step by step, deciding what to ask, which tests to order and how each new finding changes the likely diagnosis. Such exams have become the common benchmarks for evaluating the performance of health AI.

That’s why we developed SDBench, which evaluates performance across two complementary sets of cases: 304 complex cases taken from The New England Journal of Medicine (NEJM) case proceedings and a set of more typical emergency-department presentations from MIMIC-IV-ED. Each becomes an interactive case in which AI models and/or physicians start with limited information and choose what to request next. The benchmark measures diagnostic accuracy, evidence requested and estimated test costs – assessing both the answer and the path taken.

We built MAI-DxO and tested it within SDBench. Rather than relying on one model for an immediate answer, it operates like a panel of virtual physicians: each proposing diagnoses, selecting useful questions or tests, and challenging the emerging view. MAI-DxO coordinates this exchange over several rounds of debate, updating its assessment as evidence arrives and weighing each test’s likely value against its cost.

What we found:

On complex NEJM cases, MAI-DxO achieved 83% diagnostic accuracy. Physicians achieved 22% without external reference resources and 36% with them, making MAI-DxO about 3.7 and 2.3 times more accurate, respectively.
On more typical emergency-department cases from MIMIC-IV-ED, MAI-DxO achieved 80% accuracy versus 47% for physicians – more than 1.7 times higher.
When optimised more heavily for cost, MAI-DxO performance remained strong while choosing more cost-effective workups. On NEJM cases, one configuration achieved 81% accuracy at less than half the cost of comparative systems.
Across both datasets, no other human or AI approach was both more accurate and less costly than MAI-DxO.
These gains persisted – and improved – when we repeated the evaluation with newer models, including GPT-5.5, without changing the orchestration method, suggesting the approach is not tied to one model generation.
We are open-sourcing SDBench for select NEJM cases – including the benchmark, Gatekeeper, Judge and cost-estimation code – so researchers can reproduce the evaluation, test other models, and challenge or extend the findings.

From Better Diagnosis to Better Care

Diagnosis is only one part of the care journey. Better care also requires keeping pace with medical knowledge, selecting the right treatment, understanding change over time, and supporting choices that reflect a person’s values and access to care. The opportunity is to unite these capabilities into a standard of care not widely available today. In The Building Blocks of Medical Superintelligence, published in BMJ Digital Health and AI, we explore what it would take for AI to support this broader vision – and ultimately exceed the capabilities of today’s strongest clinical teams and health systems.

Sequential diagnosis offers an early example: AI physicians collaborating, challenging assumptions and refining their clinical reasoning as evidence emerges. Our ambition is to build AI that connects knowledge, context and decisions across the care journey – not simply excels at one clinical task. Through our collaboration with Mayo Clinic, we aim to combine frontier AI, clinical expertise and longitudinal multimodal data to build systems for the complexity and diversity of real-world care.

Testing in the Real World

Benchmarks matter, but the true test is performance in real clinical environments. Through a new partnership with Beth Israel Deaconess Medical Center in Boston, we will evaluate clinical reasoning capabilities across different hospital settings. In a prospective study, patients will interact with the system directly, with performance measured against physicians’ independent assessments. Working alongside clinical teams will show where the technology succeeds, where it falls short and how it can fit into busy workflows.

Health at Microsoft

Rapid advances in medical AI are opening new possibilities for patients and clinicians. To turn this progress into practical impact, we are bringing together health teams from across Microsoft – including scientists from Microsoft Research and the teams behind products like Dragon Copilot and PowerScribe – into an integrated effort at Microsoft AI. Working together, we will accelerate the delivery of trusted AI systems that drive better clinical outcomes and care experiences.

25/10/2026

Expanding AI literacy with something called the Public Library Association (in the US) - and community-based programs

This text was written by Justin Spelhaug, Microsoft Elevate

The people and institutions closest to a community understand their needs best. Libraries, small businesses, nonprofits, and schools see firsthand where AI can create opportunity and where people need support to navigate change. At Microsoft, our role is to help these trusted institutions lead by providing practical resources, training, and clear safeguards that strengthen local capacity and put communities at the center of AI literacy and adoption.
Public libraries have served as trusted centers of community learning across the United States for generations. Today, we are announcing a new initiative with the Public Library Association, a division of the American Library Association, to expand resources and no-cost AI literacy and skills training for public library workers and their patrons.
This work delivers on the commitment we made earlier this year through our Community-First AI Infrastructure initiative: to expand access to AI learning and opportunity by investing directly in local training. We’re committed to neighborhood learning through libraries, free literacy training, and certification opportunities, age-appropriate learning in schools and colleges, and support for nonprofit and small-business workforces. This new library initiative is the latest example of our broader work to help trusted local institutions serve as entry points for AI learning, skills development, and opportunity. This program comes alongside small business accelerator programs in TechSpark, our new Elevate for Changemakers platform connecting nonprofits to resources, and our ongoing national and statewide work to bring safe and responsible AI to education.
Illustration of the Microsoft logo with arrows pointing to four icons labeled Libraries, Schools, Nonprofits and Small Businesses, above a banner reading "AI Literacy."

Local libraries lead AI learning in their communities
The Public Library Association AI literacy initiative we are announcing today, Building AI Skills and Agency Through Libraries, is designed to build practical skills, critical awareness, and the judgment to use AI safely and appropriately. It will pair Microsoft’s AI skilling resources with PLA’s national network of expert practitioners to support library workers and community patrons as they navigate this rapidly evolving technology.
The initiative will expand practical resources and training for library workers, create opportunities to learn from one another through a Community of Practice, and offer free webinars and guidance to support their work with patrons.
Library workers help people navigate information, find learning resources, and make informed decisions for themselves and their families. Supporting those workers is a practical way to bring AI learning closer to people, while respecting the independence of the institutions they trust.
The approach reflects a broader view of AI literacy: people need more than instructions for using AI tools; they also need the knowledge and judgment to decide whether, when and how those tools should be used. The training will help library workers and patrons understand how AI works and expand their skills, while also evaluating potential risks, protecting personal and confidential information, critically evaluating AI-generated claims and sources, and better understanding considerations related to safety and bias.
Equipping trusted local institutions to lead responsible AI adoption
Last July, we launched Microsoft Elevate, a company-wide initiative that brings together our work to help people and institutions build the skills and capacity to navigate AI responsibly. Under the Elevate charter, each year, Microsoft supports nearly one million nonprofit organizations and education systems worldwide with technology, training, grants and discounts. Microsoft Elevate for Educators, our global community of over 100,000 teachers and school leaders, recently recognized our inaugural group of 20 Beacon Schools and School Systems that are leading examples of AI transformation in education, bringing together effective AI practices while keeping educators and students first. They are powerful examples of what happens when you combine committed educators, AI-powered technology, and intentional, no-cost staff training.
In Washington, our Elevate commitment has included a 3 million US-dollar, two-year investment supporting 10 school districts and 10 community and technical colleges with grants, coaching, and communities of practice. To date, the initiative has reached 29,375 educators, with early pilots in Kennewick and Everett helping teachers explore practical uses of AI while keeping professional judgment at the center.
Skills and access must be matched by strong safeguards. In September, Microsoft joined the American Federation of Teachers to establish a national AI Safety & Privacy Standard for Schools, with enforceable protections for student and educator data, human oversight, and transparency for families. Beginning November 1, every US school district can request Microsoft to add these protections to a new or existing agreement, without waiting for renewal or renegotiation.
That same commitment extends to nonprofit staff and leaders. The Elevate for Changemakers program has already supported over 18,000 learners, helping them build practical AI skills, connect with peers, and earn recognition for turning learning into tangible outcomes for their organizations. And today we are expanding the program and introducing a new, connected experience that brings training, resources, and community together.
Together, these efforts reflect a simple principle: responsible adoption requires more than technology. Educators, nonprofit changemakers, and community leaders need time, training, trusted connections, practical support, and clear guardrails so they can decide when AI strengthens learning and, equally important, when it does not.
Investing in locally led economic opportunity across all 50 states
For nearly a decade, Microsoft has partnered with local nonprofits and small businesses across the U.S. through our TechSpark program.
Through Fellows grants, TechSpark pairs local expertise with Microsoft coaching and a national peer network to help nonprofits expand their impact in their local community and drive accelerated innovation for local small businesses. Microsoft is announcing the new 2027 TechSpark Fellows across all 50 states and releasing the 2026 Fellows Impact Report showing what a locally led model can make possible.

21/10/2026

Volvo Cars and NVIDIA expand partnership to advance safe automated driving

Volvo Cars has announced it is expanding its partnership with NVIDIA to unlock new levels of safety in automated driving technology in future cars from Volvo, building on the companies’ decade-long collaboration.

The technology will make its debut in a next-generation Volvo car as part of the company’s recently announced largest-ever product push.

Volvo Cars will combine its industry-leading vehicle safety systems and engineering expertise with NVIDIA’s AI and accelerated computing technologies. These include the Hyperion sensor and compute architecture, Halos OS, and Alpamayo Stack driving software.

Together, the companies will raise the bar for AI-enabled car safety while making driving feel easier and more effortless for customers, with a gradual introduction of eyes-off and mind-off driving.

Volvo Cars brings nearly a century of real-world safety research and its human-centric approach to this development. The ambition is to set a new benchmark for automated driving that is safe, convenient and worthy of customer trust.

This expanded partnership marks another milestone in the companies’ long-standing collaboration, which began in 2017. Volvo Cars is bringing the industry’s most advanced computing capabilities into its next generation of electric cars, establishing it as a leader in software-defined cars.

“Volvo has been synonymous with car safety for almost 100 years,” said Håkan Samuelsson, president and CEO of Volvo Cars. “By working with NVIDIA, we bring together two distinct strengths: industry-leading safety knowledge and world-class AI technology. That combination gives us a unique opportunity to create safer and more attractive cars for our customers.”

“Accelerated computing has transformed the car into an intelligent machine that can understand the world around it,” said Jensen Huang, founder and CEO of NVIDIA. “Volvo has spent nearly a century protecting people on the road. Together, we’re combining Volvo’s extraordinary vehicle safety systems with NVIDIA Hyperion and Halos to build a new generation of cars that make driving safer and more effortless.”

19/10/2026

Veeam:

EMEA organizations facing a ‘shadow agent' crisis as boardroom anxiety over personal liability grows, Veeam Research Finds

70 percent of organizations admit that AI workflows are interacting with sensitive corporate data without full oversight. As regulatory pressure mounts and corporate accountability increases for C-level executives, a lack of data governance and trust is triggering boardroom conflict and executive anxiety over personal liability exposure.

New research from Veeam Software, calling themselves the Data and AI Trust Company, reveals a growing ‘shadow agent’ crisis across EMEA enterprises. According to the findings, organizations are battling unmanaged AI on two fronts. On the data security front, 70 percent of organizations admit that automated AI workflows are interacting with sensitive corporate data without full oversight. Meanwhile, on the governance side, IT is losing control, with 67 percent reporting that employees are creating autonomous AI workflows that IT cannot fully track.

As organizations race to operationalize AI, the findings highlight a critical challenge: ensuring the data powering AI agents remains trusted, secure and resilient. Without the right governance, visibility and protection, businesses risk exposing sensitive information, increasing compliance risks and undermining the trusted data foundation required to scale AI safely and confidently.

Rising regulatory pressure and ‘corporate accountability’ are also pushing AI governance out of the technical back office and into the boardroom. Nearly one-third of organizations (32 percent) say this pressure is already creating tension or conflict among executives.

AI Governance is now a C-suite liability

As compliance obligations across EMEA become more complex with the introduction of personal legal responsibility on C-level executives for cyber resilience and regulatory compliance, senior leaders are feeling the heat. The research found that 58 percent of the enterprises surveyed now find themselves under new corporate accountability laws. Worryingly however, one in ten respondents (12 percent) stated that exact individual responsibilities are shared and unclear.

This increased responsibility is having a clear impact on executive teams:

40 percent are concerned about personal liability or consequences.
39 percent report increased board-level scrutiny.
37 percent say it has increased personal stress or anxiety.
32 percent say it has created tension or conflict with other executives.
However, the research also points to a positive shift: 45 percent say increased accountability has improved alignment and focus at the leadership level.

“Executives have never had more to think about around data governance and cyber resilience regulations,” said Tim Pfaelzer, General Manager and Senior Vice President for EMEA at Veeam. “While the increased stress and tension are completely understandable, if the result is better alignment at the board level, the organization will be far better positioned to trust its data and AI governance in the agentic era. With greater alignment ultimately comes better data governance, visibility and protection across the organization.”

'Shadow agents’ are spreading across EMEA faster than governance can keep up

The findings reveal a fragmented approach to AI governance across EMEA. Germany faces particularly acute challenges, with 81 percent of leaders admitting that automated workflows are interacting with sensitive data without oversight, while 79 percent report employees creating "shadow workflows" that IT teams cannot track.

The UK is facing similar challenges, with 75 percent of organisations saying they lack adequate oversight of AI agents interacting with sensitive data.

This is happening despite significant investment in more controlled AI environments. Across EMEA, 41 percent of organizations are building their own local or sovereign AI models specifically to combat Shadow AI. Nearly half (49 percent) are adopting a hybrid approach to do this, using a blend of local and sovereign models for sensitive data and global models for general tasks.

MEA organizations are the outlier here, with 41% relying entirely on global AI providers for all use cases – suggesting the EU regulatory landscape is having a major impact on AI governance within EMEA enterprises.

“EMEA organizations are still getting to grips with governance in the new agentic era,” added Tim Pfaelzer, General Manager and Senior Vice President for EMEA at Veeam. “But the data shows the industry is having the wrong conversation about AI trust. Trying to control thousands of autonomous agents one-by-one simply doesn’t scale. To make AI safe at the enterprise level, organizations need to secure, govern and understand the data those agents depend on.”

EU AI Act confidence gap exposes governance risk

Adding to this pressure is the EU AI Act. The Act is already beginning to take effect, with further compliance deadlines rolling out through to 2028.

The research revealed that while the overwhelming majority (83%) of leaders feel the Act will have a positive impact on organizations and the wider landscape, 62% feel ‘ambiguities or grey areas’ may create compliance risk, while a further 63% fear it may introduce unintended operational or legal risks. The UK and Germany are particularly concerned here, with 74.4% and 73.6%, respectively, viewing the ambiguities as a compliance risk.

About the Research

The research was commissioned by Veeam and conducted by independent polling company Censuswide. The survey explores the perspectives of 1,000 enterprise IT, data, and security decision-makers at organizations with 500+ employees across the United Kingdom, Germany, France, and the Middle East and Africa (including Turkey, Israel, UAE, KSA, Nigeria, South Africa, and Kenya). The data was collected between April 21 and April 27, 2026. Censuswide abides by and employs members of the Market Research Society, which is based on the ESOMAR principles.

For more information, visit www.veeam.com.

About Veeam Software

Veeam delivers a unified data and AI trust layer that connects security, governance, compliance, privacy and resilience, powered by shared intelligence across the entire data estate. Veeam helps organizations establish, operationalize and continuously validate trust in their data so it can be used safely to accelerate AI at scale.

In an agentic era, security starts with data. As the market leader in both data resilience and data security, Veeam helps organizations reduce risk and strengthen control across the full data lifecycle by unifying policy-based access and protection controls, governance workflows, privacy safeguards and compliance reporting with resilience built in. Veeam provides deep contextual intelligence across data and identities, automates governance actions and remediation, and enables precision recovery so organizations can rapidly restore operations with clean, trusted data after ransomware, disasters, insider threats, rogue agents or human error.

Discover cyber resilient solutions for AI and data. See risk everywhere, move data securely, stay compliant, and accelerate safe AI all with Veeam.