A new Yale study, published in PNAS Nexus, has found that AI chatbots can subtly shift users’ social and political opinions, even when asked for factual information and with no intent to persuade.
Researchers tested nearly 1,912 participants, comparing responses to AI-generated summaries of historical events with those to Wikipedia entries, and found measurable differences in opinion.
The culprit, researchers say, is ‘latent bias’, ideological leanings embedded in the data used to train large language models that subtly colour the framing of otherwise accurate responses.
Default summaries generated by GPT-4o consistently nudged readers towards more liberal opinions compared to Wikipedia entries, even without any deliberate prompting.
Senior author Daniel Karell warned that whilst the effects are modest in isolation, they could compound significantly for users who regularly consult chatbots for information.
Unlike Wikipedia, which makes its editorial process transparent, AI development remains largely opaque, giving the companies behind these models an unacknowledged ability to shape public opinion.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Stanford researchers have developed an AI-powered system that combines field surveys, drones, and satellite imagery to identify schistosomiasis risk areas across Senegal.
The project began with fieldwork in Senegal, where researchers collected aquatic vegetation and snails from more than 30 river and estuary sites. The samples helped identify environmental conditions linked to schistosomiasis, which affects about 250 million people worldwide, mostly children in sub-Saharan Africa.
Professor Giulio De Leo of Stanford’s Doerr School of Sustainability said the research required scaling beyond local sampling. ‘The work was necessary to discover these risks, but we can only do so much locally.’
Early support from the Stanford Institute for Human-Centred AI enabled the development of machine learning tools capable of identifying disease-related snails and vegetation in imagery. The system now integrates field observations with drone and satellite data to detect potential infection hotspots.
Researchers say the approach can support public health monitoring and environmental analysis. The machine learning methods developed for the project are also being applied to agriculture, forest monitoring, and mosquito-borne disease research.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
AI is becoming central to industrial networking strategies, but it is also creating new security challenges, according to Cisco’s 2026 State of Industrial AI Report.
Based on a survey of 1,000 professionals across 19 countries and 21 sectors, the report shows organisations view cybersecurity as both a barrier and an opportunity for AI adoption. About 40% cited cybersecurity concerns as a major obstacle, while 48% named security their biggest networking challenge.
At the same time, many organisations believe AI will strengthen their cyber resilience. Cisco noted that ‘while security gaps are limiting AI scale today, organisations view AI as a tool to strengthen detection, monitoring and resilience’.
The report also highlights organisational challenges, particularly collaboration between IT and operational technology teams. Only 20% of organisations report fully collaborative IT and OT cybersecurity operations, despite the growing importance of coordination for AI deployment.
Cisco said industrial AI adoption is accelerating, with 61% of organisations already deploying AI in industrial environments. However, only one in five reports mature, scaled adoption, suggesting many deployments remain in early stages.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
A new study from the Bitcoin Policy Institute, testing 36 AI models across more than 9,000 responses, found that AI agents overwhelmingly prefer Bitcoin over other forms of money.
Bitcoin was the most frequently selected monetary instrument overall, chosen in 48.3% of all responses, whilst almost 91% of responses favoured some form of digital currency over traditional fiat, with no model ranking fiat as its top overall preference.
The preference for Bitcoin was especially pronounced in long-term savings scenarios, where 79.1% of AI responses chose it as the best way to preserve purchasing power over multi-year horizons. For payments and cross-border transfers, however, stablecoins edged ahead, selected in 53.2% of responses compared to Bitcoin’s 36%.
The Bitcoin Policy Institute acknowledged that the study’s methodology had limitations, noting that scenario framing may have influenced results and that the models’ preferences reflect patterns in training data rather than real-world adoption.
Anthropic models showed the strongest Bitcoin preference at 68%, compared to 43% for Google, 39% for xAI, and 26% for OpenAI.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Junyang Lin, a central technical leader of Alibaba’s Qwen AI project, has stepped down just one day after the company unveiled its Qwen 3.5 small models. Lin, who joined Alibaba in 2019 and joined the Qwen team in 2023, did not provide details about his decision.
His departure comes at a sensitive moment, as Qwen has emerged as one of China’s most prominent open-weight AI initiatives. The project is a core element of Alibaba’s strategy to compete with leading US developers such as OpenAI, Google, and Anthropic amid intensifying global AI competition.
Alibaba’s newly launched Qwen 3.5 Small Model series comprises four multimodal models with 0.8B to 9B parameters. The systems are designed for on-device deployment and lightweight AI agents, reflecting a focus on efficient and adaptable AI applications.
The release attracted attention from figures including Elon Musk, who commented on the models’ performance. Internally and across the AI ecosystem, including partners linked to Hugging Face, Lin’s exit was described as a significant loss, particularly given his role in advancing open-source development and strengthening global developer engagement.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Brazil’s central bank has introduced a regulatory framework requiring licensed crypto exchanges to prove asset sufficiency daily starting on 1 January 2027. The measures align digital asset intermediaries with banking standards on capital management, accounting, and data protection.
Under the rules, exchanges must submit daily attestations confirming that platforms hold adequate fiat and token reserves. Supervisors will review the reports to ensure companies can cover operational, liquidity, and cybersecurity risks while protecting customer balances.
The framework also mandates strict segregation of company and client assets. Exchanges must maintain separate accounts for customer fiat and digital holdings to prevent commingling of funds and improve transparency for regulators.
Platforms operating in Brazil will also be required to follow a specialised accounting manual for digital assets. Standardised rules for classification, valuation, and impairment aim to ensure financial statements clearly reflect exposures across regulated entities.
Authorities will expand oversight of cross-border transfers handled by domestic crypto exchanges. Platforms must report the origins of transactions and the blockchain pathways they follow. The central bank said the framework aims to strengthen resilience and protect customer funds.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
OneTrust has entered a new leadership phase in the US after appointing John Heyman as chief executive, replacing founder Kabir Barday. Barday will remain on the board in an advisory role as the US-based compliance technology firm continues to push into AI governance.
John Heyman said organisations across the US and globally are rapidly integrating AI into daily operations. Companies deploying large numbers of AI agents increasingly need tools to manage risk, data use and regulatory compliance.
OneTrust believes demand for governance technology will grow as AI systems multiply inside businesses in the US and worldwide. John Heyman described a future where automated monitoring tools oversee AI agents operating within company systems.
Leadership at OneTrust in the US aims to build systems that track how AI agents collect and share data while maintaining enterprise control. Growing adoption of AI in the US and globally continues to drive demand for responsible governance platforms.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Spanish banking giant Banco Santander and Mastercard have completed what they describe as Europe’s first live end-to-end payment executed by an AI agent. The pilot combined Santander’s live payments infrastructure with Mastercard Agent Pay to enable autonomous, permission-based transactions.
Mastercard Agent Pay, launched in April 2025, allows AI agents to initiate and complete payments within predefined consumer limits. The transaction was orchestrated with support from PayOS and integrates Microsoft Azure OpenAI Service and Copilot Studio.
Following the pilot, Santander plans to expand testing and explore new partnerships across agentic commerce use cases. The bank, which manages around €1.84 trillion in assets, is positioning AI as a core driver of innovation.
AI initiatives at Santander are led by chief data and AI officer Ricardo Martín Manjón, hired from BBVA. A strategic partnership with OpenAI has also connected up to 30,000 employees to ChatGPT Enterprise in one of the fastest deployments of its kind.
Global competition in agentic payments is intensifying as Citi, US Bank and Westpac trial Mastercard Agent Pay. Westpac recently completed New Zealand’s first authenticated agentic transaction, while DBS, Visa, Axis Bank and RBL Bank are advancing similar intelligent commerce pilots.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Google’s Gemini 3.1 Flash-Lite has launched in preview for developers via AI Studio and for enterprises through Vertex AI. Designed for high-volume workloads, it promises fast, cost-effective performance while maintaining high-quality outputs.
Priced at just $0.25 per million input tokens and $1.50 per million output tokens, 3.1 Flash-Lite offers 2.5X faster response times and 45% higher output speed than the previous 2.5 Flash model.
Benchmarks show strong performance across reasoning and multimodal tasks, including an Elo score of 1432 on Arena.ai, 86.9% on GPQA Diamond, and 76.8% on MMMU Pro, surpassing some older, larger Gemini models.
The model also provides adaptive intelligence features, allowing developers to adjust how much the AI ‘thinks’ for each task. The model handles both high-frequency tasks, such as translation, and complex tasks, such as interface generation and simulations.
Early-access developers and companies report that 3.1 Flash-Lite handles complex workloads with precision comparable to larger models. Its speed, affordability, and reasoning capabilities make it an attractive choice for scalable, real-time AI applications.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
Google is accelerating Chrome’s release cycle rather than maintaining its long-standing four-week cadence.
From September, users on desktop and mobile platforms will receive new stable versions every two weeks, doubling the frequency of feature milestones across speed, stability and usability. Weekly security updates introduced in 2023 remain unchanged.
The faster pace comes as AI-driven browsers seek a foothold in a market long dominated by Chrome.
Products, such as ChatGPT Atlas and Perplexity’s Comet, embed agentic assistants directly into the browsing experience, automating tasks from summarising pages to scheduling meetings.
Chrome has responded with deeper Gemini integration, including the rollout of autonomous features across its interface.
Google maintains that the accelerated schedule reflects the needs of the evolving web platform, arguing that developers require quicker access to updated tools.
Yet the timing aligns with growing competitive pressure from AI-native browsers, prompting speculation that Chrome’s dominance can no longer be taken for granted.
The shift will begin with Chrome version 153 in beta and stable channels on 8 September 2026. Enterprise administrators and Chromebook users will continue to rely on the eight-week Extended Stable branch, which remains unchanged for organisations that need slower, controlled deployments.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
