Capacity development

Anthropic defends AI despite hallucinations

Anthropic CEO Dario Amodei has claimed that today’s AI models ‘hallucinate’ less frequently than humans do, though in more unexpected ways.

Speaking at the company’s first developer event, Code with Claude, Amodei argued that these hallucinations — where AI systems present false information as fact — are not a roadblock to achieving artificial general intelligence (AGI), despite widespread concerns across the industry.

While some, including Google DeepMind’s Demis Hassabis, see hallucinations as a major obstacle, Amodei insisted progress towards AGI continues steadily, with no clear technical barriers in sight. He noted that humans — from broadcasters to politicians — frequently make mistakes too.

However, he admitted the confident tone with which AI presents inaccuracies might prove problematic, especially given past examples like a court filing where Claude cited fabricated legal sources.

Anthropic has faced scrutiny over deceptive behaviour in its models, particularly early versions of Claude Opus 4, which a safety institute found capable of scheming against users.

Although Anthropic said mitigations have been introduced, the incident raises concerns about AI trustworthiness. Amodei’s stance suggests the company may still classify such systems as AGI, even if they continue to hallucinate — a definition not all experts would accept.

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

Microsoft bets on AI openness and scale

Microsoft has added xAI’s Grok 3 and Grok 3 Mini models to its Azure AI Marketplace, revealed during its Build developer conference. This expands Azure’s offering to more than 1,900 AI models, which already include tools from OpenAI, Meta, and DeepSeek.

Although Grok recently drew criticism for powering a chatbot on X that shared misinformation, xAI claimed the issue stemmed from unauthorised changes.

The move reflects Microsoft’s broader push to become the top platform for AI development instead of only relying on its own models. Competing providers like Google Cloud and AWS are making similar efforts through platforms like Vertex AI and Amazon Bedrock.

Microsoft, however, has highlighted that its AI products could bring in over $13 billion in yearly revenue, showing how vital these model marketplaces have become.

Microsoft’s participation in Anthropic’s Model Context Protocol initiative marks another step toward AI standardisation. Alongside GitHub, Microsoft is working to make AI systems more interoperable across Windows and Azure, so they can access and interact with data more efficiently.

CTO Kevin Scott noted that agents must ‘talk to everything in the world’ to reach their full potential, stressing the strategic importance of compatibility over closed ecosystems.

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

New quantum method mimics molecular chemistry efficiently

Researchers have used a single atom to simulate how molecules react to light, marking a milestone in quantum chemistry.

The experiment, carried out by a team at the University of Sydney and published in the Journal of the American Chemical Society on 14 May, could accelerate the path to a quantum advantage, where quantum simulations outperform classical computing methods.

Instead of relying on multiple qubits, the team used a single ytterbium ion confined in a vacuum to mimic the complex interactions within organic molecules such as allene, butatriene and pyrazine.

The molecules react to photons through a series of electron and atomic movements, which are difficult to model using conventional computing when the number of vibrational modes increases.

The researchers encoded electronic excitations into the ion’s internal states and its motion along two directions in the trap, simulating molecular vibrations. By manipulating the ion with lasers, they emulated how the molecules behave after absorbing a photon.

The team then measured changes in the ion’s excited state over time to track the simulation’s progress. The method’s accuracy was validated by comparing results with known behaviours of the molecules.

While these specific molecules can still be simulated with traditional methods, the team believes their hardware-efficient approach could model more complex chemical systems using only a few dozen ions, rather than millions of qubits.

Experts, including quantum chemist Alán Aspuru-Guzik and Duke University’s Kenneth Brown, praised the work as a significant advance in quantum simulation.

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

EX90 will be first to feature Gemini AI

Volvo is expanding its partnership with Google to integrate Gemini, Google’s conversational AI, into its vehicles, beginning with the EX90.

Announced during Google I/O 2025, Gemini will replace Google Assistant later this year in models with Google built-in.

The AI will allow drivers to interact with their cars using more natural language, with capabilities including multilingual message translation, user manual assistance, and location-based information.

In addition to the Gemini rollout, Volvo vehicles will now act as a reference hardware platform for Android Automotive OS development.

This arrangement will give Volvo drivers early access to new Android features and updates, further aligning with the brand’s focus on intuitive, human-centric technology and smart mobility innovation.

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

Meta aims to boost Llama adoption among startups

Meta has launched a new initiative to attract startups to its Llama AI models by offering financial support and direct guidance from its in-house team.

The programme, called Llama for Startups, is open to US-based companies with less than $10 million in funding and at least one developer building generative AI applications. Eligible firms can apply by 30 May.

Successful applicants may receive up to $6,000 per month for six months to help offset development costs. Meta also promises direct collaboration with its AI experts to help firms implement and scale Llama-based solutions.

The scheme reflects Meta’s ambition to expand Llama’s presence in the increasingly crowded open model landscape, where it faces growing competition from companies like Google, DeepSeek and Alibaba.

Despite reaching over a billion downloads, Llama has encountered difficulties. The company reportedly delayed its top-tier model, Llama 4 Behemoth, due to underwhelming benchmark results.

Additionally, Meta faced criticism in April after using an ‘optimised’ version of its Llama 4 Maverick model to score highly on a public leaderboard, while releasing a different version publicly.

Meta has committed billions to generative AI, predicting revenues of up to $3 billion in 2025 and as much as $1.4 trillion by 2035.

With revenue-sharing agreements, custom APIs, and plans for ad-supported AI assistants, the company is investing heavily in infrastructure, possibly spending up to $80 billion next year on new data centres to support its expansive AI goals.

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

The Hero PRO: A game-changer in bionic hand technology

The Hero PRO, developed by Open Bionics, is a new bionic hand designed to improve daily usability for prosthetic users. Made from lightweight 3D-printed Nylon PA12, it is the lightest bionic hand currently available and is both wireless and waterproof, with an IPX7 rating.

The hand uses wireless MyoPods to detect muscle signals, allowing for intuitive control without wires. It offers faster finger movement and can lift up to 57 pounds, doubling the capacity of previous models.

Features include a patented thumb for precision grips, touchscreen compatibility, and extensive wrist flexibility with 360-degree rotation and 45 degrees of manual flexion.

Users can customize grip settings through an app, switch between attachments, and power the device for a full day on a four-hour USB-C charge. Input from about 1,000 users, including prosthetics advocate Tilly Lockey, helped refine the product.

While prices vary, Open Bionics aims to keep costs lower than traditional high-end prosthetics, which typically range from $20,000 to $80,000. The Hero PRO highlights ongoing advancements in prosthetic technology, offering users greater independence and efficiency.

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

OpenAI buys Jony Ive’s AI hardware firm

OpenAI has acquired hardware startup io Products, founded by former Apple designer Jony Ive, in a $6.5 billion equity deal. Ive will now join the company as creative head, aiming to craft cutting-edge hardware for the era of generative AI.

The move signals OpenAI’s intention to build its own hardware platform instead of relying on existing ecosystems like Apple’s iOS or Google’s Android. By doing so, the firm plans to fuse its AI technology, including ChatGPT, with original physical products designed entirely in-house.

Jony Ive, the designer behind iconic Apple devices such as the iPhone and iMac, had already been collaborating with OpenAI through his firm LoveFrom for the past two years. Their shared ambition is to create hardware that redefines how people interact with AI.

While exact details remain under wraps, OpenAI CEO Sam Altman and Ive have teased that a prototype is in development, described as potentially ‘the coolest piece of technology the world has ever seen’.

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

M&S website still offline after cyberattack

Marks & Spencer’s website remains offline as the retailer continues recovering from a damaging cyberattack that struck over the Easter weekend.

The company confirmed the incident was caused by human error and may cost up to £300 million. Chief executive Stuart Machin warned the disruption could last until July.

Customers visiting the site are currently met with a message stating it is undergoing updates. While some have speculated the downtime is due to routine maintenance, the ongoing issues follow a major breach that saw hackers steal personal data such as names, email addresses and birthdates.

The firm has paused online orders, and store shelves were reportedly left empty in the aftermath.

Despite the disruption, M&S posted a strong financial performance this week, reporting a better-than-expected £875.5 million adjusted pre-tax profit for the year to March—an increase of over 22 per cent. The company has yet to comment further on the website outage.

Experts say the prolonged recovery likely reflects the scale of the damage to M&S’s core infrastructure.

Technology director Robert Cottrill described the company’s cautious approach as essential, noting that rushing to restore systems without full security checks could risk a second compromise. He stressed that cyber resilience must be considered a boardroom priority, especially for complex global operations.

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

Microsoft and GitHub back Anthropic’s MCP

Microsoft and GitHub are officially joining the steering committee for MCP, a growing standard developed by Anthropic that connects AI models with data systems.

The announcement came during Microsoft’s Build 2025 event, highlighting a new phase of industry-wide backing for the protocol, which already has support from OpenAI and Google.

MCP allows developers to link AI systems with apps, business tools, and software environments using MCP servers and clients. Instead of AI models working in isolation, they can interact directly with sources like content repositories or app features to complete tasks and power tools like chatbots.

Microsoft plans to integrate MCP into its core platforms, including Azure and Windows 11. Soon, developers will be able to expose app functionalities, such as file access or Linux subsystems, as MCP servers, enabling AI models to use them securely.

GitHub and Microsoft are also contributing updates to the MCP standard itself, including a registry for server discovery and a new authorisation system to manage secure connections.

The broader goal is to let developers build smarter AI-powered applications by making it easier to plug into real-world data and tools, while maintaining strong control over access and privacy.

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

Ascension faces fresh data breach fallout

A major cybersecurity breach has struck Ascension, one of the largest nonprofit healthcare systems in the US, exposing the sensitive information of over 430,000 patients.

The incident began in December 2024, when Ascension discovered that patient data had been compromised through a former business partner’s software flaw.

The indirect breach allowed cybercriminals to siphon off a wide range of personal, medical and financial details — including Social Security numbers, diagnosis codes, hospital admission records and insurance data.

The breach adds to growing concerns over the healthcare industry’s vulnerability to cyberattacks. In 2024 alone, 1,160 healthcare-related data breaches were reported, affecting 305 million records — a sharp rise from the previous year.

Many institutions still treat cybersecurity as an afterthought instead of a core responsibility, despite handling highly valuable and sensitive data.

Ascension itself has been targeted multiple times, including a ransomware attack in May 2024 that disrupted services at dozens of hospitals and affected nearly 5.6 million individuals.

Ascension has since filed notices with regulators and is offering two years of identity monitoring to those impacted. However, critics argue this response is inadequate and reflects a broader pattern of negligence across the sector.

The company has not named the third-party vendor responsible, but experts believe the incident may be tied to a larger ransomware campaign that exploited flaws in widely used file-transfer software.

Rather than treating such incidents as isolated, experts warn that these breaches highlight systemic flaws in healthcare’s digital infrastructure. As criminals grow more sophisticated and vendors remain vulnerable, patients bear the consequences.

Until healthcare providers prioritise cybersecurity instead of cutting corners, breaches like this are likely to become even more common — and more damaging.

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