Future of work

Doctors and nurses outperform AI in patient triage

Human staff are more accurate than AI in assessing patient urgency in emergency departments, according to research presented at the European Emergency Medicine Congress in Barcelona.

The study, led by Dr Renata Jukneviciene of Vilnius University, tested ChatGPT 3.5 against clinicians and nurses using real case studies.

Doctors achieved an overall accuracy of 70.6% and nurses 65.5%, compared with 50.4% for AI. Doctors also outperformed AI in surgical and therapeutic cases, while nurses were more reliable overall.

AI did show strength in recognising the most critical cases, surpassing nurses in both accuracy and specificity. Researchers suggested that AI may help prioritise life-threatening situations and support less experienced staff instead of acting as a replacement.

However, over-triaging by AI could lead to inefficiencies, making human oversight essential.

Future studies will explore newer AI models, ECG interpretation, and integration into nurse training, particularly in mass-casualty scenarios.

Commenting on the findings, Dr Barbra Backus from Amsterdam said AI has value in certain areas, such as interpreting scans, but it cannot yet replace trained staff for triage decisions.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Claude Sonnet 4.5 expands developer options with rollbacks and longer-running agents

Anthropic has released Claude Sonnet 4.5, featuring a suite of new upgrades designed to enhance coding, automation, and creativity. The update enhances Claude Code, extends Computer Use, and introduces experimental tools to boost productivity and facilitate real-world applications.

Claude Code now features checkpoints, allowing developers to roll back projects to earlier versions. The Claude API has also been expanded, supporting longer-running agents to generate files such as slides, spreadsheets, and documents directly within chats.

The model’s Computer Use function has been strengthened, enabling agents to operate applications for up to 30 hours autonomously. Anthropic says Claude Sonnet 4.5 built a Slack-style app with 11,000 lines of code in one session.

A new feature, Imagine with Claude, focuses on generating creative software. The system produced a Shakespeare-themed desktop with customised scripts and performance schedules from a single prompt, highlighting its versatility.

Anthropic has maintained steady pricing for free and premium users, positioning Sonnet 4.5 as its most practical and feature-rich release yet, combining reliability with expanded creative and developer-friendly tools.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

Rising stress leaves cyber professionals at breaking point

Burnout is a significant challenge in the cybersecurity sector, as workers face rising threats and constant pressure to defend organisations. A BBC report highlights how professionals often feel overworked and undervalued, with stress levels leading some to take extended leave.

UK-based surveys reflect growing strain. Membership body ISC2 found that job satisfaction among cybersecurity staff dropped in 2024, with burnout cited as a key issue. Experts say demands have increased while resources remain stretched, leaving staff expected to stay on call around the clock.

Hackers are becoming more aggressive, targeting health services, retailers, and critical national infrastructure. Nation-state actors, including North Korean groups linked to large crypto thefts, are also stepping up activity. These attacks add to the psychological burden on frontline defenders.

Industry figures warn that high turnover risks weakening cyber resilience, especially in junior roles. Initiatives like Cybermindz call for better mental health support, while some argue for protections akin to those for first responders.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot! 

Harvard researchers develop AI for brain surgery

Harvard researchers have developed an AI tool to distinguish glioblastoma from similar brain tumours during surgery. The PICTURE system gives surgeons near-real-time guidance for critical decisions during surgery.

PICTURE outperformed humans and other AI, correctly distinguishing glioblastoma from PCNSL over 98 percent of the time in international tests. The tool also flags cases it is unsure of, allowing human review and reducing the risk of misdiagnosis, particularly in complex or rare brain tumours.

The AI works on frozen tissue samples, commonly used for rapid surgical evaluation, and can identify crucial cancer features such as cell shape, density, and necrosis.

Accurate tumour differentiation helps surgeons avoid unnecessary tissue removal and choose the proper treatment- surgery for glioblastoma or radiation and chemotherapy for PCNSL.

Researchers envision PICTURE could be used in surgery and pathology to aid AI collaboration, train pathologists, and improve access to neuropathology expertise. Further studies are planned to test its accuracy across more diverse populations and potentially extend its application to other cancer types.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot! 

Anthropic unveils Claude Sonnet 4.5 as the best AI coding model yet

Anthropic has released Claude Sonnet 4.5, its most advanced AI model yet, claiming state-of-the-art results in coding benchmarks. The company says the model can build production-ready applications, rather than limited prototypes, making it more reliable than earlier versions.

Claude Sonnet 4.5 is available through the Claude API and chatbot at the same price as its predecessor, with $3 per million input tokens and $15 per million output tokens.

Early enterprise tests suggest the model can autonomously code for extended periods, integrate databases, secure domains, and perform compliance checks such as SOC 2 audits.

Industry leaders have endorsed the launch, with Cursor and Windsurf calling it a new generation of AI coding models. Anthropic also emphasises more substantial alignment, noting reduced risks of deception and sycophancy, and improved resistance to prompt injection attacks.

Alongside the model, the company has introduced a Claude Agent SDK to let developers build customised agents, and launched ‘Imagine with Claude’, a research preview showing real-time code generation.

A release that highlights the intense competition in AI, with Anthropic pushing frequent updates to keep pace with rivals such as OpenAI, which has recently gained ground on coding performance with GPT-5.

Claude Sonnet 4.5 follows just weeks after Anthropic’s Claude Opus 4.1, underlining the rapid development cycles driving the sector.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

NSW expands secure AI platform NSWEduChat across schools

Following successful school trials, the New South Wales Department of Education has confirmed the broader rollout of its in-house generative AI platform, NSWEduChat.

The tool, developed within the department’s Sydney-based cloud environment, prioritises privacy, security, and equity while tailoring content to the state’s educational context. It is aligned with the NSW AI Assessment Framework.

The trial began in 16 schools in Term 1, 2024, and then expanded to 50 schools in Term 2. Teachers reported efficiency gains, and students showed strong engagement. Access was extended to all staff in Term 4, 2024, with Years 5–12 students due to follow in Term 4, 2025.

Key features include a privacy-first design, built-in safeguards, and a student mode that encourages critical thinking by offering guided prompts rather than direct answers. Staff can switch between staff and student modes for lesson planning and preparation.

All data is stored in Australia under departmental control. NSWEduChat is free and billed as the most cost-effective AI tool for schools. Other systems are accessible but not endorsed; staff must follow safety rules, while students are limited to approved tools.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

Lufthansa turns to automation and AI for efficiency

Lufthansa Group has unveiled a transformation strategy that places digitalisation and AI at the centre of its future operations. At Capital Markets Day, the company said efficiency will come from automation and streamlined processes.

Around 4,000 administrative roles are set to be cut by 2030, mainly in Germany, as Lufthansa consolidates functions and reduces duplication of work. Executives stressed that the focus will be on non-operational roles, with staff reductions to be conducted in consultation with social partners.

The airline group also confirmed continued investment in fleet renewal, with more than 230 new aircraft expected by 2030. Digital transformation and AI aim to cut costs, accelerate decisions, and boost competitiveness across the group’s airlines, cargo, and technical services.

By 2030, Lufthansa aims for an 8-10 percent EBIT margin, 15-20 percent return on capital, and over €2.5 billion in annual free cash flow. The company said these measures will ensure long-term resilience in a changing industry.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot! 

AI agents complete first secure transaction with Mastercard and PayOS

PayOS and Mastercard have completed the first live agentic payment using a Mastercard Agentic Token, marking a pivotal step for AI-driven commerce. The demonstration, powered by Mastercard Agent Pay, extends the tokenisation infrastructure that already underpins mobile payments and card storage.

The system enables AI agents to initiate payments while enforcing consent, authentication, and fraud checks, thereby forming what Mastercard refers to as the trust layer. It shows how card networks are preparing for agentic transactions to become central to digital commerce.

Mastercard’s Chief Digital Officer, Pablo Fourez, stated that the company is developing a secure and interoperable ecosystem for AI-driven payments, underpinned by tokenized credentials. The framework aims to prepare for a future where the internet itself supports native agentic commerce.

For PayOS, the milestone represents a shift from testing to commercialisation. Chief executive Johnathan McGowan said the company is now onboarding customers and offering tools for fraud prevention, payments risk management, and improved user experiences.

The achievement signals a broader transition as agentic AI moves from pilot to real-world deployment. If security models remain effective, agentic payments could soon differentiate platforms, merchants, and issuers, embedding autonomy into digital transactions.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

AI-powered Opera Neon browser launches with premium subscription

After its announcement in May, Opera has started rolling out Neon, its first AI-powered browser. Unlike traditional browsers, Neon is designed for professionals who want AI to simplify complex online workflows.

The browser introduces Tasks, which act like self-contained workspaces. AI can understand context, compare sources, and operate across multiple tabs simultaneously to manage projects more efficiently.

Neon also features cards and reusable AI prompts that users can customise or download from a community store, streamlining repeated actions and tasks.

Its standout tool, Neon Do, performs real-time on-screen actions such as opening tabs, filling forms, and gathering data, while keeping everything local. Opera says no data is shared, and all information is deleted after 30 days.

Neon is available by subscription at $19.90 per month. Invitations are limited during rollout, but Opera promises broader availability soon.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

California enacts first state-level AI safety law

In the US, California Governor Gavin Newsom has signed SB 53, a landmark law establishing transparency and safety requirements for large AI companies.

The legislation obliges major AI developers such as OpenAI, Anthropic, Meta, and Google DeepMind to disclose their safety protocols. It also introduces whistle-blower protections and a reporting mechanism for safety incidents, including cyberattacks and autonomous AI behaviour not covered by the EU AI Act.

Reactions across the industry have been mixed. Anthropic supported the law, while Meta and OpenAI lobbied against it, with OpenAI publishing an open letter urging Newsom not to sign. Tech firms have warned that state-level measures could create a patchwork of regulation that stifles innovation.

Despite resistance, the law positions California as a national leader in AI governance. Newsom said the state had demonstrated that it was possible to safeguard communities without stifling growth, calling AI ‘the new frontier in innovation’.

Similar legislation is under consideration in New York, while California lawmakers are also debating SB 243, a separate bill that would regulate AI companion chatbots.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!