The model stunned markets on its release in January and has since become Hugging Face’s most downloaded open-weight system. Unlike rivals, R1 was not trained on other models’ output but instead developed reasoning abilities through reinforcement learning.
DeepSeek’s engineers rewarded the model for correct answers, enabling it to form problem-solving strategies. Efficiency gains came from allowing R1 to score its own outputs rather than relying on a separate algorithm.
The Nature paper marks the first time a major large language model has undergone peer review. Reviewers said the process increased transparency and should be adopted by other firms as scrutiny of AI risks intensifies.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
The US Cybersecurity and Infrastructure Security Agency (CISA) has published lessons from its response to a federal agency breach.
Hackers exploited an unpatched vulnerability in GeoServer software, gaining access to multiple systems. CISA noted that the flaw had been disclosed weeks earlier and added to its Known Exploited Vulnerabilities catalogue, but the agency had not patched it in time.
Investigators also found that incident response plans were outdated and had not been tested. The lack of clear procedures delayed third-party support and restricted access to vital security tools during the investigation.
CISA added that endpoint detection alerts were not continuously reviewed and some US public-facing systems had no protection, leaving attackers free to install web shells and move laterally through the network.
The agency urged all organisations to prioritise patching, maintain and rehearse incident response plans, and ensure comprehensive logging to strengthen resilience against future cybersecurity attacks.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Avocado Consulting urges Australian organisations to boost software supply chain security after a high-alert warning from the Australian Cyber Security Centre (ACSC). The alert flagged threats, including social engineering, stolen tokens, and manipulated software packages.
Dennis Baltazar of Avocado Consulting said attackers combine social engineering with living-off-the-land techniques, making attacks appear routine. He warned that secrets left across systems can turn small slips into major breaches.
Baltazar advised immediate audits to find unmanaged privileged accounts and non-human identities. He urged embedding security into workflows by using short-lived credentials, policy-as-code, and default secret detection to reduce incidents and increase development speed for users in Australia.
Avocado Consulting advises organisations to eliminate secrets from code and pipelines, rotate tokens frequently, and validate every software dependency by default using version pinning, integrity checks, and provenance verification. Monitoring CI/CD activity for anomalies can also help detect attacks early.
Failing to act could expose cryptographic keys, facilitate privilege escalation, and result in reputational and operational damage. Avocado Consulting states that secure development practices must become the default, with automated scanning and push protection integrated into the software development lifecycle.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
A new AI system developed by the UK Cabinet Office has helped reclaim nearly £500m in fraudulent payments, marking the government’s most significant recovery of public funds in a single year.
The Fraud Risk Assessment Accelerator analyses data across government departments to identify weaknesses and prevent scams before they occur.
It uncovered unlawful council tax claims, social housing subletting, and pandemic-related fraud, including £186m linked to Covid support schemes. Ministers stated the savings would be redirected to fund nurses, teachers, and police officers.
Officials confirmed the tool will be licensed internationally, with the US, Canada, Australia, and New Zealand among the first partners expected to adopt it.
The UK announced the initiative at an anti-fraud summit with these countries, describing it as a step toward global cooperation in securing public finances through AI.
However, civil liberties groups have raised concerns about bias and oversight. Previous government AI systems used to detect welfare fraud were found to produce disparities based on age, disability, and nationality.
Campaigners warned that the expanded use of AI in fraud detection risks embedding unfair outcomes if left unchecked.
Would you like to learn more aboutAI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Meta will provide its Llama AI model to key European institutions, NATO, and several allied countries as part of efforts to strengthen national security capabilities.
The company confirmed that France, Germany, Italy, Japan, South Korea, and the EU will gain access to the open-source model. US defence and security agencies and partners in Australia, Canada, New Zealand, and the UK already use Llama.
Meta stated that the aim is to ensure democratic allies have the most advanced AI tools for decision-making, mission planning, and operational efficiency.
Although its terms bar use for direct military or espionage applications, the company emphasised that supporting allied defence strategies is in the interest of nations.
The move highlights the strategic importance of AI models in global security. Meta has positioned Llama as a counterweight to other countries’ developments, after allegations that researchers adapted earlier versions of the model for military purposes.
Would you like to learn more aboutAI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Google is rapidly expanding AI Mode, its generative AI-powered search assistant. The company has announced that the feature is now rolling out globally in Spanish. Spanish speakers can now interact with AI Mode to ask complex questions that traditional Search handles poorly.
AI Mode has seen swift adoption since its launch earlier this year. First introduced in March, the feature was rolled out to users across the US in May, followed by its first language expansion earlier this month.
Hindi, Indonesian, Japanese, Korean, and Brazilian Portuguese were the first languages added, and Spanish now joins the list. Google says more languages will follow soon as part of its global AI Mode rollout.
Google says the feature is designed to work alongside Search, not replace it, offering conversational answers with links to supporting sources. The company has stressed that responses are generated with safety filters and fact-checking layers.
The rollout reflects Google’s broader strategy to integrate generative AI into its ecosystem, spanning Search, Workspace, and Android. AI Mode will evolve with multimodal support and tighter integration with other Google services.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
ByteDance has unveiled Seedream 4.0, its latest AI-powered image generation model, which it claims outperforms Google DeepMind’s Gemini 2.5 Flash Image. The launch signals ByteDance’s bid to rival leading creative AI tools.
Developed by ByteDance’s Seed division, the model combines advanced text-to-image generation with fast, precise image editing. Internal testing reportedly showed superior prompt accuracy, image alignment, and visual quality compared to US-developed DeepMind’s system.
Artificial Analysis, an independent AI benchmarking firm, called Seedream 4.0 a significant step forward. The model integrates Seedream 3.0’s generation capability with SeedEdit 3.0’s editing tools while maintaining a price of US$30 per 1,000 generations.
ByteDance claims that Seedream 4.0 runs over 10 times faster than earlier versions, enhancing the user experience with near-instant image inference. Early users have praised its ability to make quick, text-prompted edits with high accuracy.
The tool is now available to users in China through Jimeng and Doubao AI apps and businesses via Volcano Engine, ByteDance’s cloud platform. A formal technical report supporting the company’s claims has not yet been released.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
The European Union and Indonesia have concluded negotiations on a Comprehensive Economic Partnership Agreement (CEPA) and an Investment Protection Agreement (IPA), strongly emphasising technology, digitalisation and sustainable industries.
The agreements are designed to expand trade, secure critical raw materials, and drive the green and digital transitions.
Under the CEPA, tariffs on 98.5% of trade lines will be removed, cutting costs by €600 million annually and giving EU companies greater access to Indonesia’s fast-growing technology sectors, including electric vehicles, electronics and pharmaceuticals.
European firms will also gain full ownership rights in key service areas such as computers and telecommunications, helping deepen integration of digital supply chains.
A deal that embeds commitments to the Paris Agreement while promoting renewable energy and low-carbon technologies. It also includes cooperation on digital standards, intellectual property protections and trade facilitation for sectors vital to Europe’s clean tech and digital industries.
With Indonesia as a leading producer of critical raw materials, the agreement secures sustainable and predictable access to inputs essential for semiconductors, batteries and other strategic technologies.
Launched in 2016, the negotiations concluded after the political agreement reached in July 2025 between Presidents Ursula von der Leyen and Prabowo Subianto. The texts will undergo legal review before the EU and Indonesia ratification, opening a new chapter in tech-enabled trade and innovation.
Would you like to learn more aboutAI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Google has launched Gemini for TV, bringing conversational AI to the living room. The update builds on Google TV and Google Assistant, letting viewers chat naturally with their screens to discover shows, plan trips, or even tackle homework questions.
Instead of scrolling endlessly, users can ask Gemini to find a film everyone will enjoy or recap last season’s drama. The AI can handle vague requests, like finding ‘that new hospital drama,’ and provide reviews before you press play.
Gemini also turns the TV into an interactive learning tool. From explaining why volcanoes erupt to guiding kids through projects, it offers helpful answers with supporting YouTube videos for hands-on exploration.
Beyond schoolwork, Gemini can help plan meals, teach new skills like guitar, or brainstorm family trips, all through conversational prompts. Such features make the TV a hub for entertainment, education, and inspiration.
Gemini is now available on the TCL QM9K series, with rollout to additional Google TV devices planned for later this year. Google says additional features are coming soon, making TVs more capable and personalised.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
Stellantis, the parent company of Jeep, Chrysler and Dodge, has disclosed a data breach affecting its North American customer service operations.
The company said it recently discovered unauthorised access to a third-party service platform and confirmed that customer contact details were exposed. Stellantis stressed that no financial information was compromised and that affected customers and regulators are being notified.
Cybercriminal group ShinyHunters has claimed responsibility, telling tech site BleepingComputer it had stolen over 18 million Salesforce records from the automaker, including names and contact information. Stellantis has not confirmed the number of records involved.
ShinyHunters has targeted several global firms this year, including Google, Louis Vuitton and Allianz Life, often using voice phishing to trick employees into downloading malicious software. The group claims to have stolen 1.5 billion Salesforce records from more than 700 companies worldwide.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
