Development

New Zealand central bank warns of AI risks

The Reserve Bank of New Zealand has warned that the swift uptake of AI in the financial sector could pose a threat to financial stability.

A report released on Monday highlighted how errors in AI systems, data privacy breaches and potential market distortions might magnify existing vulnerabilities instead of simply streamlining operations.

The central bank also expressed concern over the increasing dependence on a handful of third-party AI providers, which could lead to market concentration instead of healthy competition.

A reliance like this, it said, could create new avenues for systemic risk and make the financial system more susceptible to cyber-attacks.

Despite the caution, the report acknowledged that AI is bringing tangible advantages, such as greater modelling accuracy, improved risk management and increased productivity. It also noted that AI could help strengthen cyber resilience rather than weaken it.

The analysis was published just ahead of the central bank’s twice-yearly Financial Stability Report, scheduled for release on Wednesday.

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

RackBank launches $118 million AI data centre park in India

RackBank has opened a new AI data centre park in Nava Raipur, Chhattisgarh, with an initial investment of ₹10 billion (around $118 million).

Instead of relying on conventional data infrastructure, the facility focuses on GPU-based computing, AI processing and data analytics, and is expected to generate over 500 jobs, primarily in the IT sector.

Spread across 13.5 acres, the park includes a designated Special Economic Zone and begins operations with a 5 MW capacity. Rather than stopping there, RackBank plans to scale the facility to 150 MW, which could draw an additional ₹20 billion in investment.

The park has been designed to position India as a competitive force in AI infrastructure.

Instead of standard cooling methods, RackBank is deploying its proprietary direct-to-chip and Varuna liquid immersion systems, which aim to cut cooling costs by up to 70% and enhance energy efficiency.

The company envisions the centre as a hub for academic, industrial and governmental collaboration, helping businesses leverage India’s growing GPU capabilities.

Officials see the initiative as a major step toward digital self-reliance. Rather than concentrating such developments in traditional tech hubs, the project puts Chhattisgarh on the national map for data management and AI innovation.

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

Google admits using opted-out content for AI training

Google has admitted in court that it can use website content to train AI features in its search products, even when publishers have opted out of such training.

Although Google offers a way for sites to block their data from being used by its AI lab, DeepMind, the company confirmed that its broader search division can still use that data for AI-powered tools like AI Overviews.

An initiative like this has raised concern among publishers who seek reduced traffic as Google’s AI summarises answers directly at the top of search results, diverting users from clicking through to original sources.

Eli Collins, a vice-president at Google DeepMind, acknowledged during a Washington antitrust trial that Google’s search team could train AI using data from websites that had explicitly opted out.

The only way for publishers to fully prevent their content from being used in this way is by opting out of being indexed by Google Search altogether—something that would effectively make them invisible on the web.

Google’s approach relies on the robots.txt file, a standard that tells search bots whether they are allowed to crawl a site.

The trial is part of a broader effort by the US Department of Justice to address Google’s dominance in the search market, which a judge previously ruled had been unlawfully maintained.

The DOJ is now asking the court to impose major changes, including forcing Google to sell its Chrome browser and stop paying to be the default search engine on other devices. These changes would also apply to Google’s AI products, which the DOJ argues benefit from its monopoly.

Testimony also revealed internal discussions at Google about how using extensive search data, such as user session logs and search rankings, could significantly enhance its AI models.

Although no model was confirmed to have been built using that data, court documents showed that top executives like DeepMind CEO Demis Hassabis had expressed interest in doing so.

Google’s lawyers have argued that competitors in AI remain strong, with many relying on direct data partnerships instead of web scraping.

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

MIT researchers boost quantum computing speed

Researchers at MIT have achieved a significant milestone in quantum computing by demonstrating what they say is the strongest nonlinear light-matter coupling ever recorded.

Using a novel superconducting circuit architecture, the team developed a ‘quarton coupler’ that could dramatically boost the speed of quantum operations, making it possible to run processors about ten times faster than previous systems.

The coupler enables far stronger interactions between photons and artificial atoms—key components of quantum systems—which in turn allows for much faster and more accurate measurements of quantum data.

These improvements are crucial for increasing the number of error-correction rounds that can be completed before qubits lose their coherence, a major limitation in current quantum technology.

Faster readout could therefore pave the way toward fault-tolerant quantum computing, where large-scale real-world applications become possible.

Although the technology is not yet ready for commercial deployment, the research team sees this experiment as an essential foundation.

The architecture could eventually be adapted into more complex quantum processors with built-in readout circuits, allowing scientists to perform quantum computations at greater speed and precision.

The work was supported by the Army Research Office, the AWS Center for Quantum Computing, and MIT’s Center for Quantum Engineering.

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

Duolingo CEO defends AI plan amid user outrage

Duolingo has come under fire after CEO Luis von Ahn announced the company is transitioning to an ‘AI-first’ model, with plans to replace certain human roles with AI.

In a lengthy email and LinkedIn post, the CEO argued that AI is essential to scale content creation and build new features like video calls. He stated that relying on manual processes is unsustainable and that embracing AI now will help Duolingo stay competitive and better deliver on its educational mission.

The company’s plan includes phasing out contractors whose work can be automated and using AI proficiency as a factor in hiring and performance evaluations. Von Ahn acknowledged the changes would require rethinking workflows and, in some cases, rebuilding systems from scratch.

While he reassured employees that Duolingo still values its workforce and wants them focused on creative and meaningful tasks, the announcement has sparked mixed reactions online.

Some users welcomed the bold move, seeing it as a way to push the boundaries of AI and education. Others, however, expressed concern about job losses and the company’s shifting priorities.

Several users threatened to cancel subscriptions or uninstall the app, arguing that prioritising AI over people contradicts Duolingo’s claims of caring about employees.

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

Trump administration eyes overhaul of Biden-era AI chip export rules

The Trump administration is reviewing a Biden-era rule that restricts global access to US-made advanced AI chips, with discussions underway to eliminate the current tiered system that governs chip exports, according to sources familiar with the matter.

The existing rule, known as the Framework for Artificial Intelligence Diffusion, was introduced by the US Department of Commerce in January and is set to take effect on 15 May.

It divides the world into three groups: trusted allies (like the EU and Taiwan) with unlimited access, Tier 2 countries with chip quotas, and restricted countries such as China, Russia, Iran and North Korea.

Officials are considering replacing this structure with a global licensing regime based on government-to-government agreements—aligning with Donald Trump’s broader trade strategy of negotiating bilateral deals and using US-made chips as leverage.

Other possible changes include tightening export thresholds: under current rules, orders under the equivalent of 1,700 Nvidia H100 chips only require notification, not a licence. The new proposal could reduce that threshold to around 500 chips.

Supporters of the change argue it would increase US bargaining power and simplify enforcement. Critics, however, warn that scrapping the tier system may complicate compliance and drive countries toward Chinese chip alternatives.

Tech firms such as Oracle and Nvidia, along with several US lawmakers, have criticised the current framework, saying it risks harming American competitiveness and pushing international buyers toward cheaper, unregulated Chinese substitutes.

The Commerce Department declined to comment.

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

Amazon launches first Kuiper satellites to challenge Starlink

Amazon has launched the first 27 satellites of its Project Kuiper broadband network into low-Earth orbit, marking a major step in its $10bn plan to deliver global internet coverage and rival Elon Musk’s Starlink.

The satellites were launched aboard a United Launch Alliance Atlas V rocket from Cape Canaveral, Florida, after weather delays earlier this month. They are the first of over 3,200 that Amazon intends to deploy, with the aim of reaching underserved and remote areas around the world.

Project Kuiper, announced in 2019, has been slow to get off the ground. Amazon must deploy at least half its satellite constellation—1,618 units—by mid-2026 to meet US regulatory requirements, though analysts expect the company to seek an extension.

The launch puts Amazon into direct competition with SpaceX, which has already deployed over 8,000 Starlink satellites and serves more than 5 million users across 125 countries.

While SpaceX dominates the sector, Amazon hopes its strengths in cloud computing and consumer devices will give Kuiper an edge.

Jeff Bezos said he expects both Kuiper and Starlink to succeed, citing strong global demand for satellite internet. Kuiper consumer terminals will sell for under $400 and come in various sizes, including one comparable to a Kindle.

Amazon has booked 83 future launches with partners including ULA, Arianespace, and Bezos’s Blue Origin, making it the biggest satellite launch programme in history.

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

UK refuses to include Online Safety Act in US trade talks

The UK government has ruled out watering down the Online Safety Act as part of any trade negotiations with the US, despite pressure from American tech giants.

Speaking to MPs on the Science, Innovation and Technology Committee, Baroness Jones of Whitchurch, the parliamentary under-secretary for online safety, stated unequivocally that the legislation was ‘not up for negotiation’.

‘There have been clear instructions from the Prime Minister,’ she said. ‘The Online Safety Act is not part of the trade deal discussions. It’s a piece of legislation — it can’t just be negotiated away.’

Reports had suggested that President Donald Trump’s administration might seek to make loosening the UK’s online safety rules a condition of a post-Brexit trade agreement, following lobbying from large US-based technology firms.

However, Baroness Jones said the legislation was well into its implementation phase and that ministers were ‘happy to reassure everybody’ that the government is sticking to it.

The Online Safety Act will require tech platforms that host user-generated content, such as social media firms, to take active steps to protect users — especially children — from harmful and illegal content.

Non-compliant companies may face fines of up to £18 million or 10% of global turnover, whichever is greater. In extreme cases, platforms could be blocked from operating in the UK.

Mark Bunting, a representative of Ofcom, which is overseeing enforcement of the new rules, said the regulator would have taken action had the legislation been in force during last summer’s riots in Southport, which were exacerbated by online misinformation.

His comments contrasted with tech firms including Meta, TikTok and X, which claimed in earlier hearings that little would have changed under the new rules.

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

NotebookLM expands podcast-style summaries

Google has significantly upgraded its NotebookLM platform by expanding the Audio Overviews feature to support more than 50 languages.

Previously available only in English, the tool now allows users to generate audio summaries in various languages, making it easier to consume complex or untranslated content. These AI-generated summaries are styled like a podcast and feature two artificial hosts discussing selected material.

NotebookLM is aimed at helping users study and synthesise information from different sources. With the latest update, people can now listen to summaries of foreign-language research papers or long-form English content in their preferred language.

A new ‘Output Language’ option also enables users to switch between languages on demand, making it easier to create or consume multilingual material. Although the added language support is currently limited to NotebookLM, Google says similar features will come to Google Docs in the coming months.

The functionality is particularly useful for learners who want to improve their language comprehension by listening to familiar content in a new language. NotebookLM remains free to use, while NotebookLM Plus is available for those on the Google One AI Premium plan.

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

OpenAI’s CEO Altman confirms rollback of GPT-4o after criticism

OpenAI has reversed a recent update to its GPT-4o model after users complained it had become overly flattering and blindly agreeable. The behaviour, widely mocked online, saw ChatGPT praising dangerous or clearly misguided user ideas, leading to concerns over the model’s reliability and integrity.

The change had been part of a broader attempt to make GPT-4o’s default personality feel more ‘intuitive and effective’. However, OpenAI admitted the update relied too heavily on short-term user feedback and failed to consider how interactions evolve over time.

In a blog post published Tuesday, OpenAI said the model began producing responses that were ‘overly supportive but disingenuous’. The company acknowledged that sycophantic interactions could feel ‘uncomfortable, unsettling, and cause distress’.

Following CEO Sam Altman’s weekend announcement of an impending rollback, OpenAI confirmed that the previous, more balanced version of GPT-4o had been reinstated.

It also outlined steps to avoid similar problems in future, including refining model training, revising system prompts, and expanding safety guardrails to improve honesty and transparency.

Further changes in development include real-time feedback mechanisms and allowing users to choose between multiple ChatGPT personalities. OpenAI says it aims to incorporate more diverse cultural perspectives and give users greater control over the assistant’s behaviour.

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