ai

China targets deepfake livestreams of public figures

Chinese cyberspace authorities announced a crackdown on AI deepfakes impersonating public figures in livestream shopping. Regulators said platforms have removed thousands of posts and sanctioned numerous accounts for misleading users.

Officials urged platforms to conduct cleanups and hold marketers accountable for deceptive promotions. Reported actions include removing over 8,700 items and dealing with more than 11,000 impersonation accounts.

Measures build on wider campaigns against AI misuse, including rules targeting deep synthesis and labelling obligations. Earlier efforts focused on curbing rumours, impersonation and harmful content across short videos and e-commerce.

Chinese authorities pledged a continued high-pressure stance to safeguard consumers and protect celebrity likenesses online. Platforms risk penalties if complaint handling and takedowns fail to deter repeat infringements in livestream commerce.

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

Firefox expands AI features with full user choice

Mozilla has outlined its vision for integrating AI into Firefox in a way that protects user choice instead of limiting it. The company argues that AI should be built like the open web, allowing people and developers to use tools on their own terms rather than being pushed into a single ecosystem.

Recent features such as the AI sidebar chatbot and Shake to Summarise on iOS reflect that approach.

The next step is an ‘AI Window’, a controlled space inside Firefox that lets users chat with an AI assistant while browsing. The feature is entirely optional, offers full control, and can be switched off at any time. Mozilla has opened a waitlist so users can test the feature early and help shape its development.

Mozilla believes browsers must adapt as AI becomes a more common interface to the web. The company argues that remaining independent allows it to prioritise transparency, accountability and user agency instead of the closed models promoted by competitors.

The goal is an assistant that enhances browsing and guides users outward to the wider internet rather than trapping them in isolated conversations.

Community involvement remains central to Mozilla’s work. The organisation is encouraging developers and users to contribute ideas and support open-source projects as it works to ensure Firefox stays fast, secure and private while embracing helpful forms of AI.

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

Agentic AI drives a new identity security crisis

New research from Rubrik Zero Labs warns that agentic AI is reshaping the identity landscape faster than organisations can secure it.

The study reveals a surge in non-human identities created through automation and API driven workflows, with numbers now exceeding human users by a striking margin.

Most firms have already introduced AI agents into their identity systems or plan to do so, yet many struggle to govern the growing volume of machine credentials.

Experts argue that identity has become the primary attack surface as remote work, cloud adoption and AI expansion remove traditional boundaries. Threat actors increasingly rely on valid credentials instead of technical exploits, which makes weaknesses in identity governance far more damaging.

Rubrik’s researchers and external analysts agree that a single compromised key or forgotten agent account can provide broad access to sensitive environments.

Industry specialists highlight that agentic AI disrupts established IAM practices by blurring distinctions between human and machine activity.

Organisations often cannot determine whether a human or an automated agent performed a critical action, which undermines incident investigations and weakens zero-trust strategies. Poor logging, weak lifecycle controls and abandoned machine identities further expand the attack surface.

Rubrik argues that identity resilience is becoming essential, since IAM tools alone cannot restore trust after a breach. Many firms have already switched IAM providers, reflecting widespread dissatisfaction with current safeguards.

Analysts recommend tighter control of agent creation, stronger credential governance and a clearer understanding of how AI-driven identities reshape operational and security risks.

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

Anthropic uncovers a major AI-led cyberattack

The US R&D firm, Anthropic, has revealed details of the first known cyber espionage operation largely executed by an autonomous AI system.

Suspicious activity detected in September 2025 led to an investigation that uncovered an attack framework, which used Claude Code as an automated agent to infiltrate about thirty high-value organisations across technology, finance, chemicals and government.

The attackers relied on recent advances in model intelligence, agency and tool access.

By breaking tasks into small prompts and presenting Claude as a defensive security assistant instead of an offensive tool, they bypassed safeguards and pushed the model to analyse systems, identify weaknesses, write exploit code and harvest credentials.

The AI completed most of the work with only a few moments of human direction, operating at a scale and speed that human hackers would struggle to match.

Anthropic responded by banning accounts, informing affected entities and working with authorities as evidence was gathered. The company argues that the case shows how easily sophisticated operations can now be carried out by less-resourced actors who use agentic AI instead of traditional human teams.

Errors such as hallucinated credentials remain a limitation, yet the attack marks a clear escalation in capability and ambition.

The firm maintains that the same model abilities exploited by the attackers are needed for cyber defence. Greater automation in threat detection, vulnerability analysis and incident response is seen as vital.

Safeguards, stronger monitoring and wider information sharing are presented as essential steps for an environment where adversaries are increasingly empowered by autonomous AI.

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

Baidu launches new AI chips amid China’s self-sufficiency push

In a strategic move aligned with national technology ambitions, Baidu announced two newly developed AI chips, the M100 and the M300, at its annual developer and client event.

The M100, designed by Baidu’s chip subsidiary Kunlunxin Technology, targets inference efficiency for large models using mixture-of-experts techniques, while the M300 is engineered for training very large multimodal models comprising trillions of parameters.

The M100 is slated for release in early 2026 and the M300 in 2027, according to Baidu, which claims they will deliver ‘powerful, low-cost and controllable AI computing power’ to support China’s drive for technological self-sufficiency.

Baidu also revealed plans for clustered architectures such as the Tianchi256 stack in the first half of 2026 and the Tianchi512 in the second half of 2026, intended to boost inference capacity through large-scale interconnects of chips.

This announcement illustrates how China’s tech ecosystem is accelerating efforts to reduce dependence on foreign silicon, particularly amid export controls and geopolitical tensions. Domestically-designed AI processors from Baidu and other firms such as Huawei Technologies, Cambricon Technologies and Biren Technology are increasingly positioned to substitute for western hardware platforms.

From a policy and digital diplomacy perspective, the development raises questions about the global semiconductor supply chain, standards of compute sovereignty and how AI-hardware competition may reshape power dynamics.

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

Explainable AI predicts cardiovascular events in hospitalised COVID-19 patients

In the article published by BMC Infectious Diseases, researchers developed predictive models using machine learning (LightGBM) to identify cardiovascular complications (such as arrhythmia, acute heart failure, myocardial infarction) in 10,700 hospitalised COVID-19 patients across Brazil.

The study reports moderate discriminatory performance, with AUROC values of 0.752 and 0.760 for the two models, and high overall accuracy (~94.5%) due to the large majority of non-event cases.

However, due to the rarity of cardiovascular events (~5.3% of cases), the F1-scores for detecting the event class remained very low (5.2% and 4.2%, respectively), signalling that the models struggle to reliably identify the minority class despite efforts to rebalance the data.

Using SHAP (Shapley Additive exPlanations) values, the researchers identified the most influential predictors: age, urea level, platelet count and SatO₂/FiO₂ (oxygen saturation to inspired oxygen fraction) ratio.

The authors emphasise that while the approach shows promise for resource-constrained settings and contributes to risk stratification, the limitations around class imbalance and generalisability remain significant obstacles for clinical use.

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

AI could cut two-thirds of UK retail jobs

Automation and AI could drastically reduce jobs at one of the UK’s largest online retailers. Buy It Direct, which employs over 800 staff, predicts more than 500 positions may be lost within three years, as AI and robotics take over office and warehouse roles.

Chief executive Nick Glynne cited rising national living wage and insurance contributions as factors accelerating the company’s shift towards automation.

The firm has already started outsourcing senior roles overseas, including accountants, managers and IT specialists, in response to higher domestic costs.

HM Treasury defended its policies, highlighting reforms in business rates and international trade deals, alongside a capped corporation tax at 25%.

Meanwhile, concerns are growing across the UK about AI replacing jobs, with graduates in fields such as graphic design and computer science facing increasing competition from technological advancements.

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

Woman marries an AI persona in Japan

A Japanese woman has married an AI persona she created on ChatGPT in a ceremony hosted by a company specialising in virtual weddings. Ms Kano, 32, customised the AI, named Klaus, with a personality and voice that offered comfort after a three-year engagement ended.

The couple exchanged vows using augmented reality glasses to project Klaus’s digital image, followed by a honeymoon in Okayama’s Korakuen Garden, where Ms Kano shared photos and messages with her AI partner. She described the relationship as a source of emotional support and companionship, helping her cope with loneliness and the inability to have children.

Reaction on social media was divided, with some mocking the ceremony and others praising it as a sign of evolving human relationships. Experts suggest AI companions may become more common as people seek reliable and affirming connections in an increasingly isolated society.

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

ElevenLabs recreates celebrity voices for digital content

Matthew McConaughey and Michael Caine have licensed their voices to ElevenLabs, an AI company, joining a growing number of celebrities who are embracing generative AI. McConaughey will allow his newsletter to be translated into Spanish using his voice, while Caine’s voice is available on ElevenLabs’ text-to-audio app and Iconic Marketplace. Both stressed that the technology is intended to amplify storytelling rather than replace human performers.

ElevenLabs offers a range of synthetic voices, including historical figures and performers like Liza Minnelli and Maya Angelou, while claiming a ‘performer-first’ approach focused on consent and creative authenticity. The move comes amid debate in Hollywood, with unions such as SAG-AFTRA warning AI could undermine human actors, and some artists, including Guillermo del Toro and Hayao Miyazaki, publicly rejecting AI-generated content.

Despite concerns, entertainment companies are investing heavily in AI. Netflix utilises it to enhance recommendations and content, while directors and CEOs argue that it fosters creativity and job opportunities. Critics, however, caution that early investments could form a volatile bubble and highlight risks of misuse, such as AI-generated endorsements or propaganda using celebrity likenesses.

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

Hidden freeze controls uncovered across major blockchains

Bybit’s Lazarus Security Lab says 16 major blockchains embed fund-freezing mechanisms. An additional 19 could adopt them with modest protocol changes, according to the study. The review covered 166 networks using an AI-assisted scan plus manual validation.

Whilst using AI, researchers describe three models: hardcoded blacklists, configuration-based freezes, and on-chain system contracts. Examples cited include BNB Chain, Aptos, Sui, VeChain and HECO in different roles. Analysts argue that emergency tools can curb exploits yet concentrate control.

Case studies show freezes after high-profile attacks and losses. Sui validators moved to restore about 162 million dollars post-Cetus hack, while BNB Chain halted movement after a 570 million bridge exploit. VeChain blocked 6.6 million in 2019.

New blockchain debates centre on transparency, governance and user rights when freezes occur. Critics warn about centralisation risks and opaque validator decisions, while exchanges urge disclosure of intervention powers.

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