Artificial intelligence

Claude AI gains powerful file editing tools for documents and spreadsheets

Anthropic’s Claude has expanded its role as a leading AI assistant by adding advanced tools for creating and editing files. Instead of manually working with different programs, users can now describe their needs in plain language and let the AI produce or update Word, Excel, PowerPoint, and PDF files.

A feature that supports uploads of CSV and TSV data and can generate charts, graphs, or images where needed, with a 30MB size limit applying to uploads and downloads.

The real breakthrough lies in editing. Instead of opening a document or spreadsheet, users can simply type instructions such as replacing text, changing currencies, or updating job titles. Claude processes the prompt and makes all the changes in one pass, preserving the original formatting.

It positions the AI as more efficient than rivals, as Gemini can only export reports but not directly modify existing files.

The feature preview is available on web and desktop for subscribers on Max, Team, or Enterprise plans. Analysts suggest the update could reshape productivity tools, especially after reports that Microsoft has partnered with Anthropic to explore using Claude for Office 365 functions.

By removing repetitive tasks and making file handling conversational, Claude is pushing productivity software into a new phase of automation.

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

Small business revival could hinge on AI-driven tools

If AI is to matter in the economy, it must first matter to small businesses. Firms employ over 61 million people, nearly half the private workforce, yet most run on outdated technology. While smartphones update monthly, many small businesses still use systems built a decade ago.

Search fund entrepreneurs bridge this gap by upgrading established firms with modern tech. One deal turned a 50-person roadside assistance firm into Asurion, now a global tech-care provider. Others have scaled compliance firms into nationwide SaaS platforms.

Generative AI now accelerates these transformations, cutting work times by over 60% across supply chains, compliance, and document processing functions. Complex tasks can now be completed in hours, unlocking double-digit productivity gains and allowing small businesses to focus on growth.

Search funds are not the only path forward. AI consulting firms, tech studios, and AI-powered roll-up strategies bring enterprise-grade tools to family-run firms. For communities that have relied on traditional playbooks, decades of growth can be compressed into months.

The cost of AI has never been lower, and the opportunity is wide open. Once deployed at scale, AI could power a wave of productivity on Main Street, helping small businesses compete and strengthening the economy for half of their workforce.

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

AI and AFM deliver real-time macrophage phenotyping

Macrophages drive immune responses, including inflammation, tissue repair, and tumour growth. Identifying their polarisation states is key for diagnosis and immunotherapy, but current methods, such as RNA sequencing and flow cytometry, are expensive, slow, and unsuitable for real-time use.

Atomic force microscopy (AFM) has emerged as a powerful tool for decoding mechanobiological signatures of cells. Combined with AI, AFM data can be rapidly analysed, but macrophage phenotyping has been relatively underexplored using this approach.

Researchers led by Prof Li Yang at the Shenzhen Institutes of Advanced Technology have now developed a label-free, non-invasive method combining AFM with deep learning. The system accurately profiles human macrophage mechanophenotypes and identifies polarisation states in real-time.

The AI model was trained on well-characterised macrophage subtypes and validated using flow cytometry. Results showed that pseudovirus stimulation mainly produced M1 macrophages, with smaller populations of M2 and mixed phenotypes, closely matching the model’s predictions.

The study, published in Small Methods, offers a promising diagnostic tool that could be extended beyond macrophages to other cell types. It could support new approaches in cancer, fibrosis, and infectious disease diagnostics based on mechanophenotypes.

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

NotebookLM turns notes into flashcards podcasts and quizzes

Google’s learning-focused AI tool NotebookLM has gained a major update, making studying and teaching more interactive.

Instead of offering only static summaries, it now generates flashcards that condense key information into easy-to-remember notes, helping users recall knowledge more effectively.

Reports can also be transformed into quizzes with customisable topics and difficulty, which can then be shared with friends or colleagues through a simple link.

The update extends to audio learning, where NotebookLM’s podcast-style Audio Overviews are evolving with new formats. Instead of a single style, users can now create Brief, Debate, or Critique episodes, giving greater flexibility in how material is explained or discussed.

Google is also strengthening its teaching tools. A new Blog Post format offers contextual suggestions such as strategy papers or explainers, while the ability to create custom report formats allows users to design study resources tailored to their needs.

The most significant addition, however, is the Learning Guide. Acting like a personal tutor, it promotes deeper understanding by asking open-ended questions, breaking problems into smaller steps, and adapting explanations to suit each learner.

With these features, NotebookLM is moving closer to becoming a comprehensive learning assistant, offering a mix of interactive study aids and adaptable teaching methods that go beyond simple note-taking.

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

Google Cloud scraps transfer fees to win multicloud users

Google Cloud has rolled out a new Data Transfer Essentials plan, allowing customers to move workloads across multiple clouds without paying transfer fees. The move comes ahead of the EU Data Act on 12 September 2025, intended to strengthen competition across the cloud market.

Under the new rules, providers must charge data transfer fees only ‘at cost.’ While Microsoft and Amazon have already taken steps to comply, Google has gone further by removing the charge altogether.

The company said the change is aimed at supporting businesses with multicloud strategies, offering more flexibility, and reducing downtime in critical workloads.

The initiative also positions Google as more aligned with regulatory goals, particularly compared with Microsoft, which has faced scrutiny over restrictive licensing practices. Google said qualifying traffic will now be billed at zero cost, while other transfers remain charged at existing rates.

The announcement follows strong growth in Google Cloud’s business, especially from AI firms like OpenAI and Anthropic. Alphabet’s cloud contracts are valued at around $106 billion, with CEO Thomas Kurian projecting $58 billion in revenue conversion within two years.

Alphabet’s stock price rose 2.47% following the update, reaching $239.94, as investors responded positively to both growth prospects and regulatory positioning.

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

Photonic chips open the path to sustainable AI by training with light

A team of international researchers has shown how training neural networks directly with light on photonic chips could make AI faster and more sustainable.

A breakthrough study, published in Nature, involved collaboration between the Politecnico di Milano, EPFL Lausanne, Stanford University, the University of Cambridge, and the Max Planck Institute.

The research highlights how physical neural networks, which use analogue circuits that exploit the laws of physics, can process information in new ways.

Photonic chips developed at the Politecnico di Milano perform mathematical operations such as addition and multiplication through light interference on silicon microchips only a few millimetres in size.

By eliminating the need to digitise information, these chips dramatically cut both processing time and energy use. Researchers have also pioneered an ‘in-situ’ training technique that enables photonic neural networks to learn tasks entirely through light signals, instead of relying on digital models.

The result is a training process that is faster, more efficient and more robust.

Such advances could lead to more powerful AI models capable of running directly on devices instead of being dependent on energy-hungry data centres.

An approach that paves the way for technologies such as autonomous vehicles, portable intelligent sensors and real-time data processing systems that are both greener and quicker.

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

Oracle and OpenAI drive record $300B investment in cloud for AI

OpenAI has finalised a record $300 billion deal with Oracle to secure vast computing infrastructure over five years, marking one of the most significant cloud contracts in history. The agreement is part of Project Stargate, OpenAI’s plan to build massive data centre capacity in the US and abroad.

The two companies will develop 4.5 gigawatts of computing power, equivalent to the energy consumed by millions of homes.

Backed by SoftBank and other partners, the Stargate initiative aims to surpass $500 billion in investment, with construction already underway in Texas. Additional plans include a large-scale data centre project in the United Arab Emirates, supported by Emirati firm G42.

The scale of the deal highlights the fierce race among tech giants to dominate AI infrastructure. Amazon, Microsoft, Google and Meta are also pledging hundreds of billions of dollars towards data centres, while OpenAI faces mounting financial pressure.

The company currently generates around $10 billion in revenue but is expected to spend far more than that annually to support its expansion.

Oracle is betting heavily on OpenAI as a future growth driver, although the risk is high given OpenAI’s lack of profitability and Oracle’s growing debt burden.

A gamble that rests on the assumption that ChatGPT and related AI technologies will continue to grow at an unprecedented pace, despite intense competition from Google, Anthropic and others.

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

Cybersecurity protections for US companies at risk as key law nears expiration

As cyber threats grow, a vital legal safeguard encouraging US companies to share threat intelligence is on the verge of expiring.

The US Cybersecurity Information Sharing Act of 2015 (CISA 2015), which grants liability protection to firms that voluntarily share cyber threat data with peers and the federal government, is set to lapse at the end of the month unless Congress acts swiftly.

The potential loss of this law could leave companies, especially small and mid-sized organisations, isolated in defending against cyberattacks, including those powered by emerging technologies like agentic AI. Companies may revert to lengthy legal reviews without liability protection or avoid information-sharing altogether.

On 3 September 2025, the House Homeland Security Committee unanimously approved a bill to extend these protections, but it still needs full congressional approval and the president’s signature.

According to Bloomberg, the Cybersecurity and Infrastructure Security Agency (CISA) has suffered budget cuts and workforce reductions under the Trump administration. Despite the administration’s criticism of the agency, its nominee to lead CISA, Sean Plankey, has publicly supported extending CISA 2015.

Industry leaders warn that losing these protections could slow down vital threat coordination. ‘This is the last line of defence,’ said Carole House, a former White House cybersecurity advisor.

With the potential expiration of CISA 2015, industry-focused Information Sharing and Analysis Centres (ISACs), now numbering at least 28 in the USA, may serve as a fallback for cybersecurity collaboration.

While some ISACs already offer legal protections like NDAs and anonymous sharing, experts warn that companies may hesitate to participate without federal liability protections.

Complex legal agreements could become necessary, potentially limiting engagement. ‘You run the risk of some companies deciding it’s too risky,’ said Scott Algeier, executive director of the IT-ISAC, despite hopes for continued collaboration.

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

Growing concern over AI fatigue among students and teachers

Experts say growing exposure to AI is leaving many people exhausted, a phenomenon increasingly described as ‘AI fatigue’.

Educators and policymakers note that AI adoption surged before society had time to thoroughly weigh its ethical or social effects. The technology now underpins tasks from homework writing to digital art, leaving some feeling overwhelmed or displaced.

University students are among those most affected, with many relying heavily on AI for assignments. Teachers say it has become challenging to identify AI-generated work, as detection tools often produce inconsistent results.

Some educators are experimenting with low-tech classrooms, banning phones and requiring handwritten work. They report deeper conversations and stronger engagement when distractions are removed.

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

AI chatbot mixes up Dutch political party policies

Dutch voters have been warned not to rely on AI chatbots for political advice after Google’s NotebookLM mixed up VVD and PVV policies.

When asked about Ukrainian refugees, the tool attributed a PVV proposal to send men back to Ukraine to the VVD programme. Similar confusions reportedly occurred when others used the system.

Google acknowledged the mistake and said it would investigate whether the error was a hallucination, the term for incorrect AI-generated output.

Experts caution that language models predict patterns rather than facts, making errors unavoidable. Voting guide StemWijzer stressed that reliable political advice requires up-to-date and verified information.

Professor Claes de Vreese said chatbots might be helpful to supplementary tools but should never replace reading actual party programmes. He also urged stricter regulation to avoid undue influence on election choices.

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