Capacity development

User emails and phone numbers leaked in Substack security incident

Substack confirmed a data breach that exposed user email addresses and phone numbers. The company said passwords and financial information were not affected. The incident occurred in October and was later investigated.

Chief executive Chris Best told users the vulnerability was identified in February and has since been fixed, with an internal investigation now underway. The company has not disclosed the technical cause of the breach or why the intrusion went undetected for several months.

Substack also did not confirm how many users were affected or provide evidence showing whether the exposed data has been misused. Users were advised to remain cautious about unexpected emails and text messages following the incident.

The breach was first reported by TechCrunch, which said the company declined to provide further operational details. Questions remain around potential ransom demands or broader system access.

Substack reports more than 50 million active subscriptions, including 5 million paid users, and raised $100 million in Series C funding in 2025, led by BOND and The Chernin Group, with participation from Andreessen Horowitz and other investors.

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

MIT researchers unveil EnCompass for AI agent search

Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory and Asari AI have introduced EnCompass, a framework designed to enhance how AI agents interact with large language models.

The system improves agent performance by automatically backtracking when errors occur and running multiple execution paths in parallel to identify the most effective outcome.

Programming AI agents traditionally requires extensive additional code to handle model mistakes. EnCompass removes that burden by embedding retry and search logic directly into execution.

Developers annotate key decision points, allowing the framework to explore alternative reasoning paths while preserving the agent’s original workflow structure.

Efficiency gains appear significant. Trials show coding effort for search implementation reduced by as much as 80%, while accuracy in code translation tasks improved between 15% and 40%.

Researchers demonstrated the framework’s ability to optimise repository translation and rule discovery across complex digital systems.

Future applications extend to large-scale software maintenance, scientific experimentation, and engineering design. Presented at NeurIPS, EnCompass positions structured search as key to advancing reliable, high-performance AI agent systems.

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

Claude Opus 4.6 sets new benchmark for enterprise AI

Anthropic has released Claude Opus 4.6, its most advanced AI model to date, introducing significant improvements in coding performance, reasoning depth, and long-context comprehension.

Engineering workflows stand to benefit from stronger debugging, code review, and better large-scale repository management, while agentic task execution now runs for more extended periods with greater reliability.

The model’s 1M token context window, now in beta, enables sustained reasoning across vast datasets and extended conversations. Performance gains span multiple benchmarks, leading in agentic coding, multidisciplinary reasoning, and high-value knowledge work.

Information retrieval in long documents has also improved, addressing persistent industry concerns around context degradation.

Operational capabilities extend beyond software development into enterprise productivity. Financial analysis, research, and document workflows gain direct support, with spreadsheet and presentation integrations enhancing daily business use.

Within Claude Code, newly introduced agent teams allow multiple AI agents to collaborate autonomously on complex workloads.

Safety remains central, with expanded evaluations showing low misalignment risk backed by interpretability research and cybersecurity safeguards. Additional tools- adaptive thinking, effort scaling, and context compaction- add flexibility for deploying long-running AI systems at scale.

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

Frontier and GPT-5.3-Codex mark major OpenAI expansion

OpenAI has unveiled Frontier, a new enterprise platform designed to help organisations build, deploy, and manage AI agents capable of executing real operational work.

The launch reflects accelerating enterprise adoption, with businesses reporting measurable productivity gains across manufacturing, finance, sales, and energy operations through agent deployment.

Frontier addresses a growing gap between AI model capability and real-world implementation. The platform equips AI agents with shared organisational context, system access, governance permissions, and feedback learning mechanisms.

By integrating across existing cloud infrastructure, enterprise software, and data environments, Frontier enables AI coworkers to operate across workflows rather than within isolated tools.

Alongside the platform release, OpenAI introduced GPT-5.3-Codex, its most advanced agentic coding model to date. The system combines the coding strength of earlier Codex iterations with expanded reasoning and professional task execution.

Benchmark performance leads across SWE-Bench Pro, Terminal-Bench, OSWorld, and GDPval, reflecting gains in software engineering, computer use, and knowledge work automation.

Cybersecurity capabilities also advance with the release. GPT-5.3-Codex includes enhanced vulnerability detection training and operates under strengthened safeguards designed to support defensive research while mitigating misuse.

Together, Frontier and GPT-5.3-Codex position AI agents as scalable digital coworkers capable of executing complex technical and enterprise workloads end-to-end.

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

AI predicts brain age and cancer survival

Researchers at Mass General Brigham have unveiled BrainIAC, an artificial intelligence model capable of analysing brain MRI scans to predict age, dementia risk, tumour mutations, and cancer survival. The model demonstrates remarkable flexibility, handling a wide variety of medical tasks with high accuracy.

BrainIAC employs self-supervised learning to identify features from unlabeled MRI datasets, allowing it to adapt to numerous clinical applications without requiring extensive annotated data. Its performance surpasses that of conventional task-specific AI frameworks.

Tests on nearly 49,000 MRI scans across seven different tasks revealed the model’s ability to generalise across both healthy and abnormal images. It successfully tackled both straightforward tasks, such as scan classification, and complex challenges, including tumour mutation detection.

The team highlights BrainIAC’s potential to accelerate biomarker discovery, improve diagnostic tools, and personalise patient care. While results are promising, researchers note that further studies on additional imaging techniques and larger datasets are necessary to validate its broader clinical use.

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

Startup founded by Nobel laureate focuses on scalable quantum chips

Renowned physicist John Martinis, a Nobel Prize winner, is pursuing a new quantum computing breakthrough. His early work proved electrical circuits could behave like quantum particles, enabling modern quantum machines.

Momentum grew when Martinis led Google’s ‘quantum supremacy’ experiment, outperforming classical computers in specialised tasks. Scaling remains difficult because fragile qubits, complex wiring and manufacturing limits reduce reliability.

Startup QoLab, founded in 2024, is redesigning quantum chip architecture to solve those hardware problems. Integrating components onto chips could reduce wiring, improve stability and enable larger systems.

Useful quantum computers could transform chemistry, materials science and complex simulations beyond classical limits. Martinis believes hardware innovation and scalable manufacturing will determine future industry leaders.

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

AI set to redefine Indian industry and economy

Artificial intelligence is becoming a cornerstone of India’s economic and industrial growth. The upcoming AI summit highlights the goal of building AI as national infrastructure, reflecting India’s languages, values, and knowledge systems.

Indian IT and service industries are moving beyond software maintenance to providing AI infrastructure and intelligent systems. Such a transformation can automate workflows, boost productivity, and create new opportunities domestically and globally.

Industrial AI is set to transform manufacturing, enabling next-generation factories through virtual twin technologies. AI grounded in physics and industrial knowledge allows faster prototyping, efficient resource use, and greater competitiveness for large enterprises and MSMEs.

Collaborations between NVIDIA and Dassault Systèmes showcase AI-driven factories and industrial intelligence. India’s talent, scale, and digital ecosystem position it to lead in industrial and generative AI, setting global technological and economic benchmarks.

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

New AI system predicts iceberg movements to aid maritime safety

Scientists are applying AI to enhance iceberg tracking and movement prediction, using machine learning models that analyse historical drift patterns, ocean currents and weather data.

These AI systems aim to identify how icebergs behave over time and improve forecasts for their positions, which can help maritime operators and climate researchers anticipate hazards and plan safe routes.

Traditional methods for tracking icebergs, relying on satellite imagery and manual analysis, are limited by coverage gaps and delays. The new AI techniques can fill these gaps by generating continuous trajectory predictions, enabling more proactive monitoring over remote polar waters.

Researchers suggest that this approach could support shipping safety, offshore operations, and environmental management as climate change alters iceberg calving and drift behaviours.

This work reflects broader trends in using AI for environmental modelling, where machine learning augments physical models to better understand complex natural systems influenced by changing climate conditions.

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

AI-powered translation glasses unlock powerful campus access

Pusan National University is piloting AI-powered translation glasses that display real-time subtitles of Korean-language lectures, aiming to reduce language barriers for international students. As a result, students can follow classes more easily, grasp specialised terminology, and engage more fully without the constant risk of missing key points.

In addition to academic settings, the technology is improving communication across campus life. For example, university staff, including counsellors, report that the glasses enable more natural, face-to-face conversations with foreign students, rather than relying on phones or other intermediary devices for translation.

Moreover, the pilot supports a broader push to internationalise the campus through AI-based multilingual services, including translated course syllabi and websites, with wider rollout to follow pending evaluation.

At the same time, the company behind the glasses is looking to expand adoption beyond Busan. If deployed more widely, the technology could influence higher education policies by easing language requirements and helping universities attract more international students, particularly as domestic enrolment declines.

However, several practical challenges remain. While translation accuracy is already high, issues such as device weight and battery life have prompted the development of lighter models. As the system continues to be refined and trained on academic vocabulary, its reliability and usability are expected to improve further.

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

Agent-based automation in Claude Cowork sparks fears of a SaaS disruption wave

Anthropic has expanded its AI assistant Claude with Cowork, an agent-based workspace for everyday office tasks. Users can grant controlled folder access so Claude can create, edit, and organise files within workflows. Cowork is available in research preview for Claude Max subscribers on macOS.

Claude Cowork breaks work into step-by-step plans and carries out tasks independently. Multiple jobs can run at once, from sorting documents to producing reports from notes or screenshots. The system is positioned as a digital colleague rather than a chatbot.

Anthropic has introduced 11 plug-ins that extend Claude Cowork across legal, sales, marketing, support, and data analysis. Organisations can define workflows, apply brand rules, and integrate business data into task execution. The tools are designed to be customisable without technical complexity.

The company has open-sourced its initial plug-ins and expects enterprises to build tailored versions. Previously part of Claude Code, the tools are now integrated into Claude Cowork through a simplified interface. Anthropic frames the update as embedding AI directly into operations.

Market reaction has highlighted fears that agent-based AI could disrupt software services. Major IT stocks in India reportedly fell following the launch. The term ‘SaaSpocalypse’ reflects unease about AI becoming core infrastructure.

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