DeepSeek claims its new AI rivals top global models

Chinese AI start-up DeepSeek has announced a major upgrade to its R1 model, stating that the new version, R1-0528, now matches the performance of top-tier models from global giants like OpenAI and Google.

Unveiled on Thursday, the upgraded model brings notable improvements in reasoning and creative writing, allowing it to generate argumentative essays, fiction, and prose with a style said to be close to human authors.

DeepSeek also reported enhanced coding capabilities, positioning the model as a more versatile tool for both technical and literary applications.

The original R1 model debuted in January. With the R1-0528 update, DeepSeek appears to be accelerating its efforts to compete with established Western AI developers in both domestic and international markets.

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

Gemini AI can now summarise videos in Google Drive

Google is expanding Gemini AI’s capabilities in Drive by enabling it to analyse video files and respond to user questions or generate concise summaries.

The new feature aims to save users time by providing quick insights from lengthy content such as meetings, classes or announcements, instead of requiring them to watch the entire video. Until now, Gemini could only summarise documents and PDFs stored in Drive.

According to a blog post published on 28 May 2025, the feature will support prompts like ‘Summarise the video’ or ‘List action items from the meeting.’ Users can access Gemini’s functionality either through Drive’s overlay previewer or a standalone viewer in a separate browser tab.

However, captions must be enabled within the user’s domain for the feature to work properly.

The update is being gradually rolled out and is expected to be available to all eligible users by 19 June. At the moment, it is limited to English and accessible only to users of Google Workspace and Google One AI Premium, or those with Gemini Business or Enterprise add-ons.

For administrators, smart features and personalisation settings must be activated to grant access.

To use the new function, users can double-click on a video file in Drive and select the ‘Ask Gemini’ option marked by a star icon in the top right corner. Google says the upgrade reflects a broader effort to integrate AI seamlessly into everyday workflows by making content easier to navigate and understand.

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

Meta faces backlash over open source AI claims

Meta is under renewed scrutiny for what critics describe as ‘open washing’ after sponsoring a Linux Foundation whitepaper on the benefits of open source AI.

The paper highlights how open models help reduce enterprise costs—claiming companies using proprietary AI tools spend over three times more. However, Meta’s involvement has raised questions, as its Llama AI models are presented as open source despite industry experts insisting otherwise.

Amanda Brock, head of OpenUK, argues that Llama does not meet accepted definitions of open source due to licensing terms that restrict commercial use.

She referenced the Open Source Initiative’s (OSI) standards, which Llama fails to meet, pointing to the presence of commercial limitations that contradict open source principles. Brock noted that open source should allow unrestricted use, which Llama’s license does not support.

Meta has long branded its Llama models as open source, but the OSI and other stakeholders have repeatedly pushed back, stating that the company’s licensing undermines the very foundation of open access.

While Brock acknowledged Meta’s contribution to the broader open source conversation, she also warned that such mislabelling could have serious consequences—especially as lawmakers and regulators increasingly reference open source in crafting AI legislation.

Other firms have faced similar allegations, including Databricks with its DBRX model in 2024, which was also criticised for failing to meet OSI standards. As the AI sector continues to evolve, the line between truly open and merely accessible models remains a point of growing tension.

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

Anthropic CEO warns of mass job losses from AI

Just one week after releasing its most advanced AI models to date — Opus 4 and Sonnet 4 — Anthropic CEO Dario Amodei warned in an interview with Axios that AI could soon reshape the job market in alarming ways.

AI, he said, may be responsible for eliminating up to half of all entry-level white-collar roles within the next one to five years, potentially driving unemployment as high as 10% to 20%.

Amodei’s goal in speaking publicly is to help workers prepare and to urge both AI companies and governments to be more transparent about coming changes. ‘Most of them [workers] are unaware that this is about to happen,’ he told Axios. ‘It sounds crazy, and people just don’t believe it.’

According to Amodei, the shift from AI augmenting jobs to fully automating them could begin as soon as two years from now. He highlighted how widespread displacement may threaten democratic stability and deepen inequality, as large groups of people lose the ability to generate economic value.

Despite these warnings, Amodei explained that competitive pressures prevent developers from slowing down. Regulatory caution in the US, he suggested, would only result in countries like China advancing more rapidly.

Still, not all implications are negative. Amodei pointed to major breakthroughs in other areas, such as healthcare, as part of the broader impact of AI.

‘Cancer is cured, the economy grows at 10% a year, the budget is balanced — and 20% of people don’t have jobs,’ he said.

To prepare society, Amodei called for increased public awareness, encouraging individuals to reconsider career paths and avoid the most automation-prone fields.

He referenced the Anthropic Economic Index, which monitors how AI affects different occupations. At its launch in February, the index showed that 57% of AI use cases still supported human tasks rather than replacing them.

However, during a press-only session at Code with Claude, Amodei noted that augmentation is likely to be a short-term strategy. He described a ‘rising waterline’ — the gradual shift from assistance to full replacement — which may soon outpace efforts to retain human roles.

‘When I think about how to make things more augmentative, that is a strategy for the short and the medium term — in the long term, we are all going to have to contend with the idea that everything humans do is eventually going to be done by AI systems. That is a constant. That will happen,’ he said.

His other recommendations included boosting AI literacy and equipping public officials with a deeper understanding of superintelligent systems, so they can begin forming policy for a radically transformed economy.

While Amodei’s outlook may sound daunting, it echoes a pattern seen throughout history: every major technological disruption brings workforce upheaval. Though some roles vanish, others emerge. Several studies suggest AI may even highlight the continued relevance of distinctively human skills.

Regardless of the outcome, one thing remains clear — learning to work with AI has never been more important.

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

New York Times partners with Amazon on AI integration

The New York Times Company and Amazon have signed a multi-year licensing agreement that will allow Amazon to integrate editorial content from The New York Times, NYT Cooking, and The Athletic into a range of its AI-powered services, the companies announced Wednesday.

Under the deal, Amazon will use licensed content for real-time display in consumer-facing products such as Alexa, as well as for training its proprietary foundation models. The agreement marks an expansion of the firms’ existing partnership.

‘The agreement expands the companies’ existing relationship, and will deliver additional value to Amazon customers while bringing Times journalism to broader audiences,’ the companies said in a joint statement.

According to the announcement, the licensing terms include ‘real-time display of summaries and short excerpts of Times content within Amazon products and services’ alongside permission to use the content in AI model development. Amazon platforms will also feature direct links to full Times articles.

Both companies described the partnership as a reflection of a shared commitment to delivering global news and information across Amazon’s AI ecosystem. Financial details of the agreement were not made public.

The announcement comes amid growing industry debate about the role of journalistic material in training AI systems.

By entering a formal licensing arrangement, The New York Times positions itself as one of the first major media outlets to publicly align with a technology company for AI-related content use.

The companies have yet to name additional Amazon products that will feature Times content, and no timeline has been disclosed for the rollout of the new integrations.

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

EU says US tech firms censor more

Far more online content is removed under US tech firms’ terms and conditions than under the EU’s Digital Services Act (DSA), according to Tech Commissioner Henna Virkkunen.

Her comments respond to criticism from American tech leaders, including Elon Musk, who have labelled the DSA a threat to free speech.

In an interview with Euractiv, Virkkunen said recent data show that 99% of content removals in the EU between September 2023 and April 2024 were carried out by platforms like Meta and X based on their own rules, not due to EU regulation.

Only 1% of cases involved ‘trusted flaggers’ — vetted organisations that report illegal content to national authorities. Just 0.001% of those reports led to an actual takedown decision by authorities, she added.

The DSA’s transparency rules made those figures available. ‘Often in the US, platforms have more strict rules with content,’ Virkkunen noted.

She gave examples such as discussions about euthanasia and nude artworks, which are often removed under US platform policies but remain online under European guidelines.

Virkkunen recently met with US tech CEOs and lawmakers, including Republican Congressman Jim Jordan, a prominent critic of the DSA and the DMA.

She said the data helped clarify how EU rules actually work. ‘It is important always to underline that the DSA only applies in the European territory,’ she said.

While pushing back against American criticism, Virkkunen avoided direct attacks on individuals like Elon Musk or Mark Zuckerberg. She suggested platform resistance reflects business models and service design choices.

Asked about delays in final decisions under the DSA — including open cases against Meta and X — Virkkunen stressed the need for a strong legal basis before enforcement.

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

AT&T hit by alleged 31 million record breach

A hacker has allegedly leaked data from 31 million AT&T customers, raising fresh concerns over the security of one of America’s largest telecom providers. The data, posted on a major dark web forum in late May 2025, is said to contain 3.1GB of customer information in both JSON and CSV formats.

Instead of isolated details, the breach reportedly includes highly sensitive data: full names, dates of birth, tax IDs, physical and email addresses, device and cookie identifiers, phone numbers, and IP addresses.

Cybersecurity firm DarkEye flagged the leak, warning that the structured formats make the data easy for criminals to exploit.

If verified, the breach would mark yet another major incident for AT&T. In March 2024, the company confirmed that personal information from 73 million users had been leaked.

Just months later, a July breach exposed call records and location metadata for nearly 110 million customers, with blame directed at compromised Snowflake cloud accounts.

AT&T has yet to comment on the latest claims. Experts warn that the combination of tax numbers and device data could enable identity theft, financial scams, and advanced phishing attacks.

For a company already under scrutiny for past security lapses, the latest breach could further damage public trust.

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

Chinese tech firms thrive amid US curbs

Nvidia CEO Jensen Huang has raised concerns that US export restrictions are accelerating the growth of Chinese AI firms, making them more competitive. In a recent interview, Huang highlighted that companies like Huawei — long blacklisted by the US — have become ‘formidable’ rivals.

The restrictions have hit Nvidia hard, with the company projecting an $8 billion revenue loss this quarter due to the limited access to the Chinese market, historically its largest for chips. The Biden administration’s AI diffusion rule, aimed at controlling the spread of advanced AI technologies, had already drawn criticism from Huang before being partially rolled back by the Trump administration in May.

Yet, Nvidia still faces tough restrictions, including a ban on selling even its downgraded H20 chip to China. The company was recently notified that it would need a special license to export the chip, leaving it with no viable alternative for the Chinese market.

Huang warned that efforts to keep cutting-edge AI tech out of China have largely backfired, as Chinese firms are finding workarounds and quickly catching up.

‘They’re doubling or quadrupling capabilities every year,’ he noted, stressing that the performance of Huawei’s latest AI chip now rivals Nvidia’s once-leading H200.

Despite Nvidia’s strong recent performance, Huang emphasised the long-term importance of re-engaging with China, home to the world’s largest community of AI researchers. He urged US policymakers to reconsider their approach, advocating for broader access to American AI technology to maintain leadership and influence in the global AI ecosystem.

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

Victoria’s Secret website hit by cyber attack

Victoria’s Secret’s website has remained offline for three days due to a security incident the company has yet to fully explain. A spokesperson confirmed steps are being taken to address the issue, saying external experts have been called in and some in-store systems were also taken down as a precaution.

Instead of revealing specific details, the retailer has left users with only a holding message on a pink background. It has declined to comment on whether ransomware is involved, when the disruption began, or if law enforcement has been contacted.

The firm’s physical stores continue operating as normal, and payment systems are unaffected, suggesting the breach has hit other digital infrastructure. Still, the shutdown has rattled investors—shares fell nearly seven percent on Wednesday.

With online sales accounting for a third of Victoria’s Secret’s $6 billion annual revenue, the pressure to resolve the situation is high.

The timing has raised eyebrows, as cybercriminals often strike during public holidays like Memorial Day, when IT teams are short-staffed. The attack follows a worrying trend among retailers.

UK giants such as Harrods, Marks & Spencer, and the Co-op have all suffered recent breaches. Experts warn that US chains are becoming the next major targets, with threat groups like Scattered Spider shifting their focus across the Atlantic.

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

Google Drive adds AI video summaries

Google Drive is gaining a new AI-powered tool that allows Workspace users to summarise and interact with video content using Gemini, Google’s generative AI assistant.

Instead of manually skipping through videos, users can now click the ‘Ask Gemini’ button to get instant summaries, key highlights, or action items from uploaded recordings.

The tool builds on Gemini 2.5 Pro’s strong video analysis capabilities, which recently scored 84.8% on the VideoMME benchmark. Gemini’s side panel, already used for summarising documents and folders, can now handle natural language prompts like ‘Summarise this video’ or ‘List key points from this meeting’.

However, the feature only works in English and requires captions to be enabled by the Workspace admin.

Google is rolling out the feature across various Workspace plans, including Business Standard and Enterprise tiers, with access available through Drive’s overlay preview or a new browser tab.

Instead of switching between windows or scrubbing through videos, users can now save time by letting Gemini handle the heavy lifting.

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