Sustainable development

Gemini 2.5 Pro tops AI coding tests, surpasses ChatGPT and Claude

Google has released an updated version of its Gemini 2.5 Pro model, addressing issues found in earlier updates.

Unlike the I/O Edition, which focuses mostly on coding, the new version improves performance more broadly and is expected to become a stable release in both the Gemini app and web interface.

The company claims the updated model performs significantly better in code generation, topping the Aider Polyglot test with a score of 82.2 percent—surpassing offerings from OpenAI, Anthropic and DeepSeek.

Beyond coding, the model aims to close previous performance gaps introduced with the March 25th update, especially in creativity and response formatting.

Developers can now fine-tune the model’s ‘thinking budget’, while users should notice a more transparent output structure. These changes and consistent improvement in leaderboard ratings on LMArena and WebDevArena suggest that Google is extending its lead in the AI race.

Google continues to rely on blind testing to judge how people feel about its models, and the new Gemini Pro seems to resonate well. In fact, it now answers even quirky test questions with more clarity and confidence—something that had been lacking in earlier versions.

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

Reddit accuses Anthropic of misusing user content

Reddit has taken legal action against AI startup Anthropic, alleging that the company scraped its platform without permission and used the data to train and commercialise its Claude AI models.

The lawsuit, filed in San Francisco’s Superior Court, accuses Anthropic of breaching contract terms, unjust enrichment, and interfering with Reddit’s operations.

According to Reddit, Anthropic accessed the platform more than 100,000 times despite publicly claiming to have stopped doing so.

The complaint claims Anthropic ignored Reddit’s technical safeguards, such as robots.txt files, and bypassed the platform’s user agreement to extract large volumes of user-generated content.

Reddit argues that Anthropic’s actions undermine its licensing deals with companies like OpenAI and Google, who have agreed to strict content usage and deletion protocols.

The filing asserts that Anthropic intentionally used personal data from Reddit without ever seeking user consent, calling the company’s conduct deceptive. Despite public statements suggesting respect for privacy and web-scraping limitations, Anthropic is portrayed as having disregarded both.

The lawsuit even cites Anthropic’s own 2021 research that acknowledged Reddit content as useful in training AI models.

Reddit is now seeking damages, repayment of profits, and a court order to stop Anthropic from using its data further. The market responded positively, with Reddit’s shares closing nearly 67% higher at $118.21—indicating investor support for the company’s aggressive stance on data protection.

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

OpenAI and India plan AI infrastructure push

OpenAI is in discussions with the Indian government to collaborate on data centre infrastructure as part of its new global initiative, ‘OpenAI for Countries’.

The programme aims to help partner nations expand AI capabilities through joint investment and strategic coordination with the US. India could become one of the ten initial countries in the effort, although specific terms remain under wraps.

During a visit to Delhi, OpenAI’s chief strategy officer Jason Kwon emphasised India’s potential, citing the government’s clear focus on infrastructure and AI talent.

Similar to the UAE’s recently announced Stargate project in Abu Dhabi, India may host large-scale AI computing infrastructure while also investing in the US under the same framework.

To nurture AI skills, OpenAI and the Ministry of Electronics and IT’s IndiaAI Mission launched the ‘OpenAI Academy’. It marks OpenAI’s first international rollout of its educational platform.

The partnership will provide free access to AI tools, developer training, and events, with content in English, Hindi, and four additional regional languages. It will also support government officials and startups through dedicated learning platforms.

The collaboration includes hackathons, workshops in six cities, and up to $100,000 in API credits for selected IndiaAI fellows and startups. The aim is to accelerate innovation and help Indian developers and researchers scale AI solutions more efficiently, according to IT Minister Ashwini Vaishnaw.

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

China accuses Taiwan of cyber attacks and offers a bounty

Authorities in Guangzhou have placed a secret bounty on more than 20 individuals suspected of launching cyber attacks on Chinese targets, according to state news agency Xinhua.

One named suspect, Ning Enwei, is reportedly linked to Taiwan’s government. While the size of the reward remains undisclosed, officials claim the accused hackers targeted sectors including defence, aerospace, energy, and science—alongside agencies in Hong Kong and Macau.

Xinhua stated that Taiwan’s ‘information, communication and digital army’ has coordinated with US forces to carry out cyber and cognitive warfare against China.

These accusations form part of a broader Chinese narrative suggesting Taiwan is seeking independence through foreign alliances, particularly with US intelligence agencies. State media also claimed the US has trained Taiwanese personnel and helped orchestrate cyber attacks on the mainland.

In response, a senior Taiwanese security official, speaking anonymously, dismissed the claims as fabricated. The official argued that Beijing is attempting to deflect criticism following allegations of Chinese cyber activities in Europe, especially in the Czech Republic.

‘It is typical of the Chinese Communist Party’s efforts to change the narrative,’ the official said, branding Beijing an international cyber threat instead of a victim.

Taiwan’s government has yet to issue an official statement.

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

Amazon invests $10 billion in AI data centres

Amazon is investing $10 billion to build data centres in North Carolina, aiming to expand its AI and cloud infrastructure instead of outsourcing compute needs.

The initiative will create at least 500 high-skilled roles and includes support for education, broadband careers, and local development through a $150,000 community fund.

The company is also developing AI-powered humanoid robots for future delivery tasks, reportedly testing them in a newly constructed ‘humanoid park’ at its San Francisco office.

Although using third-party hardware for now, the long-term goal is to embed Amazon’s software into these robots, according to sources.

Experts say the investment underlines a growing concern: that only large firms can afford the infrastructure needed for cutting-edge AI.

‘It’s positive for growth but risks concentrating innovation in Big Tech’s hands,’ said Leo Fan, co-founder of Cysic, a blockchain-based AI firm. He argues that the shift could disincentivise smaller players and dampen broader AI progress.

Amazon is also rolling out Alexa+, a new generative AI-powered assistant, and has secured a licensing deal with The New York Times to integrate journalism, recipes, and sports content into Alexa and other AI products, further extending its ecosystem instead of relying solely on internal data.

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

OpenAI hits 3 million business subscribers

OpenAI has added another 1 million paying business subscribers since February, bringing the total to 3 million across ChatGPT Enterprise, Team and Edu.

The milestone was shared during a company livestream and confirmed in interviews with outlets like CNBC.

Chief Operating Officer Brad Lightcap noted that the business tools are being adopted widely, even in regulated sectors like finance and healthcare.

He said growth among individual users has fuelled enterprise adoption instead of stalling it, highlighting a feedback loop between consumer and business uptake.

OpenAI launched ChatGPT Enterprise in August 2023, followed by Team in January 2024 and Edu in May 2024. Within a year of its first business product, the firm had already reached 1 million paying business users—a number that has now tripled.

Lightcap said AI is reshaping work across sectors—from student learning to patient care and public services—by increasing productivity instead of just automating tasks.

A separate PYMNTS Intelligence report found that 82% of workers using generative AI weekly believe it improves their output. OpenAI’s overall user base has reportedly reached 800 million people, with CEO Sam Altman claiming 10% of the global population now uses the company’s tools.

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

Google email will reply by using your voice

Google is building a next-generation email system that uses generative AI to reply to mundane messages in your own tone, according to DeepMind CEO Demis Hassabis.

Speaking at SXSW London, Hassabis said the system would handle everyday emails instead of requiring users to write repetitive responses themselves.

Hassabis called email ‘the thing I really want to get rid of,’ and joked he’d pay thousands each month for that luxury. He emphasised that while AI could help cure diseases or combat climate change, it should also solve smaller daily annoyances first—like managing inbox overload.

The upcoming feature aims to identify routine emails and draft replies that reflect the user’s writing style, potentially making decisions on simpler matters.

While details are still limited, the project remains under development and could debut as part of Google’s premium AI subscription model before reaching free-tier users.

Gmail already includes generative tools that adjust message tone, but the new system goes further—automating replies instead of just suggesting edits.

Hassabis also envisioned a universal AI assistant that protects users’ attention and supports digital well-being, offering personalised recommendations and taking care of routine digital tasks.

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

Salt Typhoon and Silk Typhoon reveal weaknesses

Recent revelations about Salt Typhoon and Silk Typhoon have exposed severe weaknesses in how organisations secure their networks.

These state-affiliated hacking groups have demonstrated that modern cyber threats come from well-resourced and coordinated actors instead of isolated individuals.

Salt Typhoon, responsible for one of the largest cyber intrusions into US infrastructure, exploited cloud network vulnerabilities targeting telecom giants like AT&T and Verizon, forcing companies to reassess their reliance on traditional private circuits.

Many firms continue to believe private circuits offer better protection simply because they are off the public internet. Some even add MACsec encryption for extra defence. However, MACsec’s ‘hop-by-hop’ design introduces new risks—data is repeatedly decrypted and re-encrypted at each routing point.

Every one of these hops becomes a possible target for attackers, who can intercept, manipulate, or exfiltrate data without detection, especially when third-party infrastructure is involved.

Beyond its security limitations, MACsec presents high operational complexity and cost, making it unsuitable for today’s cloud-first environments. In contrast, solutions like Internet Protocol Security (IPSec) offer simpler, end-to-end encryption.

Although not perfect in cloud settings, IPSec can be enhanced through parallel connections or expert guidance. The Cybersecurity and Infrastructure Security Agency (CISA) urges organisations to prioritise complete encryption of all data in transit, regardless of the underlying network.

Silk Typhoon has further amplified concerns by exploiting privileged credentials and cloud APIs to infiltrate both on-premise and cloud systems. These actors use covert networks to maintain long-term access while remaining hidden.

As threats evolve, companies must adopt Zero Trust principles, strengthen identity controls, and closely monitor their cloud environments instead of relying on outdated security models.

Collaborating with cloud security experts can help shut down exposure risks and protect sensitive data from sophisticated and persistent threats.

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

HMRC got targeted in a £47 million UK fraud

A phishing scheme run by organised crime groups cost the UK government £47 million, according to officials from His Majesty’s Revenue and Customs.

Criminals posed as taxpayers to claim payments using fake or hijacked credentials. Rather than a cyberattack, the operation relied on impersonation and did not involve the theft of taxpayer data.

Angela MacDonald, HMRC’s deputy chief executive, confirmed to Parliament’s Treasury Committee that the fraud took place in 2024. The stolen funds were taken through three separate payments, though HMRC managed to block an additional £1.9 million attempt.

Officials began a cross-border criminal investigation soon after discovering the scam, which has led to arrests.

Around 100,000 PAYE accounts — typically used by employers for employee tax and national insurance payments — were either created fraudulently or accessed illegally.

Banks were also targeted through the use of HMRC-linked identity information. Customers first flagged the issue when they noticed unusual activity.

HMRC has shut down the fake accounts and removed false data as part of its response. John-Paul Marks, HMRC’s chief executive, assured the committee that the incident is now under control and contained. ‘That is a lot of money and unacceptable,’ MacDonald told MPs.

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

Meta inks 20-year nuclear deal to power AI expansion

Meta has entered a landmark 20-year agreement with Constellation to purchase 1.1 gigawatts of nuclear power from the Clinton Clean Energy Center in Illinois, starting in 2027.

The deal is designed to support the company’s rapidly growing AI infrastructure and data centres as energy demands surge across the tech industry.

Once facing closure due to financial losses, the Clinton plant’s future is secure — without relying on Illinois’ Zero Emission Credit programme. The agreement will keep over 1,100 local jobs, boost grid capacity by 30 megawatts, and generate an estimated $13.5 million in annual tax revenue.

Illinois lawmakers have praised the deal for its economic and environmental benefits, with Republican Regan Deering calling it ‘a forward-thinking investment.’

The partnership is part of Meta’s broader strategy to build a nuclear-powered AI ecosystem. With clean energy targets of 1 to 4 gigawatts, Meta has been negotiating with multiple nuclear providers and says further agreements are in the final stages.

According to the International Atomic Energy Agency, global data centre energy use is set to more than double by 2030 — potentially outstripping Japan’s entire electricity consumption. Meta alone plans to invest $65 billion in AI infrastructure in 2025.

The Clinton plant deal also serves as a hedge against the environmental impact of fossil fuels. A 2024 study by the Brattle Group estimated that closing the facility would have led to an additional 34 million metric tons of carbon emissions over two decades. It would also have dealt an annual $765 million blow to Illinois’ GDP.

Constellation, the plant’s operator, said consistent, carbon-free baseload power is essential for the AI-driven future. With its reliability and scale, nuclear energy is increasingly seen as critical to supporting always-on AI systems.

Meanwhile, Meta continues advancing its AI vision. The company plans to fully automate ad creation by late 2026, generating images, videos, and text tailored to user location and timing.

This automation effort has already boosted ad performance, with Q1 2025 results showing a 30% rise in AI-generated ad use, a 10% increase in average ad prices, and $42.31 billion in revenue — a 16% year-over-year jump.

However, the push for AI-generated content has unsettled the advertising industry. Firms like Omnicom Group have seen share prices dip over fears disrupting to traditional creative and production models.

Zuckerberg’s long-term AI vision includes automating marketing and enhancing user experience through AI companions and virtual therapists — part of Meta’s goal to integrate machine learning into everyday life while ensuring its platforms run on clean, scalable energy.

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