Access

UK report quantifies rapid advances in frontier AI capabilities

For the first time, the UK has published a detailed, evidence-based assessment of frontier AI capabilities. The Frontier AI Trends Report draws on two years of structured testing across areas including cybersecurity, software engineering, chemistry, and biology.

The findings show rapid progress in technical performance. Success rates on apprentice-level cyber tasks rose from under 9% in 2023 to around 50% in 2025, while models also completed expert-level cyber challenges previously requiring a decade of experience.

Safeguards designed to limit misuse are also improving, according to the report. Red-team testing found that the time required to identify universal jailbreaks increased from minutes to several hours between model generations, representing an estimated forty-fold improvement in resistance.

The analysis highlights advances beyond cybersecurity. AI systems now complete hour-long software engineering tasks more than 40% of the time, while biology and chemistry models outperform PhD-level researchers in controlled knowledge tests and support non-experts in laboratory-style workflows.

While the report avoids policy recommendations, UK officials say it strengthens transparency around advanced AI systems. The government plans to continue investing in evaluation science through the AI Security Institute, supporting independent testing and international collaboration.

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

Strong AI memory demand boosts Micron outlook into 2026

Micron Technology reported record first-quarter revenue for fiscal 2026, supported by strong pricing, a favourable product mix and operating leverage. The company said tight supply conditions and robust AI-related demand are expected to continue into 2026.

The Boise-based chipmaker generated $13.64 billion in quarterly revenue, led by record sales across DRAM, NAND, high-bandwidth memory and data centres. Chief executive Sanjay Mehrotra said structural shifts are driving rising demand for advanced memory in AI workloads.

Margins expanded sharply, setting Micron apart from peers such as Broadcom and Oracle, which reported margin pressure in recent earnings. Chief financial officer Mark Murphy said gross margin is expected to rise further in the second quarter, supported by higher prices, lower costs and a favourable revenue mix.

Analysts highlighted improving fundamentals and longer-term visibility. Baird said DRAM and NAND pricing could rise sequentially as Micron finalises long-term supply agreements, while capital expenditure plans for fiscal 2026 were viewed as manageable and focused on expanding high-margin HBM capacity.

Retail sentiment also turned strongly positive following the earnings release, with Micron shares jumping around 8 per cent in after-hours trading. The stock is on track to finish the year as the best-performing semiconductor company in the S&P 500, reinforcing confidence in its AI-driven growth trajectory.

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

Natural language meets robotics in MIT’s on-demand object creation system

MIT researchers have developed a speech-to-reality system that allows users to create physical objects by describing them aloud, combining generative AI with robotic assembly. The system can produce simple furniture and decorative items in minutes using modular components.

The workflow translates spoken instructions into a digital design using a large language model and 3D generative AI. The design is then broken into voxel-based parts and adapted to real-world fabrication constraints before being assembled by a robotic arm.

Researchers have demonstrated the system by producing stools, shelves, chairs, tables and small sculptures. The approach aims to reduce manufacturing complexity by enabling rapid construction without specialised knowledge of 3D modelling or robotics.

Unlike traditional fabrication methods such as 3D printing, which can take hours or days, the modular assembly process operates quickly and allows objects to be disassembled and reused. The team is exploring stronger connection methods and extensions to larger-scale robotic systems.

The research was presented at the ACM Symposium on Computational Fabrication in November. The team said the work points toward more accessible, flexible and sustainable ways to produce physical objects using natural language and AI-driven design.

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

New Kimwolf Android botnet linked to a record-breaking DDoS attacks

Cybersecurity researchers have uncovered a rapidly expanding Android botnet known as Kimwolf, which has already compromised approximately 1.8 million devices worldwide.

The malware primarily targets smart TVs, set-top boxes, and tablets connected to residential networks, with infections concentrated in countries including Brazil, India, the US, Argentina, South Africa, and the Philippines.

Analysis by QiAnXin XLab indicates that Kimwolf demonstrates a high degree of operational resilience.

Despite multiple disruptions to its command-and-control infrastructure, the botnet has repeatedly re-emerged with enhanced capabilities, including the adoption of Ethereum Name Service to harden its communications against takedown efforts.

Researchers also identified significant similarities between Kimwolf and AISURU, one of the most powerful botnets observed in recent years. Shared source code, infrastructure, and infection scripts suggest both botnets are operated by the same threat group and have coexisted on large numbers of infected devices.

AISURU has previously drawn attention for launching record-setting distributed denial-of-service attacks, including traffic peaks approaching 30 terabits per second.

The emergence of Kimwolf alongside such activity highlights the growing scale and sophistication of botnet-driven cyber threats targeting global internet infrastructure.

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

PwC automates AI governance with Agent Mode

The global professional services network, PwC, has expanded its Model Edge platform with the launch of Agent Mode, an AI assistant designed to automate governance, compliance and documentation across enterprise AI model lifecycles.

The capability targets the growing administrative burden faced by organisations as AI model portfolios scale and regulatory expectations intensify.

Agent Mode allows users to describe governance tasks in natural language, instead of manually navigating workflows.

A system that executes actions directly within Model Edge, generates leadership-ready documentation and supports common document and reporting formats, significantly reducing routine compliance effort.

PwC estimates weekly time savings of between 20 and 50 percent for governance and model risk teams.

Behind the interface, a secure orchestration engine interprets user intent, verifies role based permissions and selects appropriate large language models based on task complexity. The design ensures governance guardrails remain intact while enabling faster and more consistent oversight.

PwC positions Agent Mode as a step towards fully automated, agent-driven AI governance, enabling organisations to focus expert attention on risk assessment and regulatory judgement instead of process management as enterprise AI adoption accelerates.

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

Healthcare faces growing compliance pressure from AI adoption

AI is becoming a practical tool across healthcare as providers face rising patient demand, chronic disease and limited resources.

These AI systems increasingly support tasks such as clinical documentation, billing, diagnostics and personalised treatment instead of relying solely on manual processes, allowing clinicians to focus more directly on patient care.

At the same time, AI introduces significant compliance and safety risks. Algorithmic bias, opaque decision-making, and outdated training data can affect clinical outcomes, raising questions about accountability when errors occur.

Regulators are signalling that healthcare organisations cannot delegate responsibility to automated systems and must retain meaningful human oversight over AI-assisted decisions.

Regulatory exposure spans federal and state frameworks, including HIPAA privacy rules, FDA oversight of AI-enabled medical devices and enforcement under the False Claims Act.

Healthcare providers are expected to implement robust procurement checks, continuous monitoring, governance structures and patient consent practices as AI regulation evolves towards a more coordinated national approach.

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

US platforms signal political shift in DSA risk reports

Major online platforms have submitted their 2025 systemic risk assessments under the Digital Services Act as the European Commission moves towards issuing its first fine against a Very Large Online Platform.

The reports arrive amid mounting political friction between Brussels and Washington, placing platform compliance under heightened scrutiny on both regulatory and geopolitical fronts.

Several US-based companies adjusted how risks related to hate speech, misinformation and diversity are framed, reflecting political changes in the US while maintaining formal alignment with EU law.

Meta softened enforcement language, reclassified hate speech under broader categories and reduced visibility of civil rights structures, while continuing to emphasise freedom of expression as a guiding principle.

Google and YouTube similarly narrowed references to misinformation, replaced established terminology with less charged language and limited enforcement narratives to cases involving severe harm.

LinkedIn followed comparable patterns, removing references to earlier commitments on health misinformation, civic integrity and EU voluntary codes that have since been integrated into the DSA framework.

X largely retained its prior approach, although its report continues to reference cooperation with governments and civil society that contrasts with the platform’s public positioning.

TikTok diverged from other platforms by expanding disclosures on hate speech, election integrity and fact-checking, likely reflecting its vulnerability to regulatory action in both the EU and the US.

European regulators are expected to assess whether these shifts represent genuine risk mitigation or strategic alignment with US political priorities.

As systemic risk reports increasingly inform enforcement decisions, subtle changes in language, scope and emphasis may carry regulatory consequences well beyond their formal compliance function.

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

OpenAI adds pinned chat feature to ChatGPT apps

The US tech company, OpenAI, has begun rolling out a pinned chats feature in ChatGPT across web, Android and iOS, allowing users to keep selected conversations fixed at the top of their chat history for faster access.

The function mirrors familiar behaviour from messaging platforms such as WhatsApp and Telegram instead of requiring repeated scrolling through past chats.

Users can pin a conversation by selecting the three-dot menu on the web or by long-pressing on mobile devices, ensuring that essential discussions remain visible regardless of how many new chats are created.

An update that follows earlier interface changes aimed at helping users explore conversation paths without losing the original discussion thread.

Alongside pinned chats, OpenAI is moving ChatGPT toward a more app-driven experience through an internal directory that allows users to connect third-party services directly within conversations.

The company says these integrations support tasks such as bookings, file handling and document creation without switching applications.

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

Instacart faces FTC scrutiny over AI pricing tool

US regulators are examining Instacart’s use of AI in grocery pricing, after reports that shoppers were shown different prices for identical items. Sources told Reuters the Federal Trade Commission has opened a probe into the company’s AI-driven pricing practices.

The FTC has issued a civil investigative demand seeking information about Instacart’s Eversight tool, which allows retailers to test different prices using AI. The agency said it does not comment on ongoing investigations, but expressed concern over reports of alleged pricing behaviour.

Scrutiny follows a study of 437 shoppers across four US cities, which found average price differences of 7 percent for the same grocery lists at the same stores. Some shoppers reportedly paid up to 23 percent more than others for identical items, according to the researchers.

Instacart said the pricing experiments were randomised and not based on personal data or individual behaviour. The company maintains that retailers, not Instacart, set prices on the platform, with the exception of Target, where prices are sourced externally and adjusted to cover costs.

The investigation comes amid wider regulatory focus on technology-driven pricing as living costs remain politically sensitive in the United States. Lawmakers have urged greater transparency, while the FTC continues broader inquiries into AI tools used to analyse consumer data and set prices.

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

ChatGPT expands with a new app directory from OpenAI

OpenAI has opened submissions for third-party apps inside ChatGPT, allowing developers to publish tools that extend conversations with real-world actions. Approved apps will appear in a new in-product directory, enabling users to move directly from discussion to execution.

The initiative builds on OpenAI’s earlier DevDay announcement, where it outlined how apps could add specialised context to conversations. Developers can now submit apps for review, provided they meet the company’s requirements on safety, privacy, and user experience.

ChatGPT apps are designed to support practical workflows such as ordering groceries, creating slide decks, or searching for apartments. Apps can be activated during conversations via the tools menu, by mentioning them directly, or through automated recommendations based on context and usage signals.

To support adoption, OpenAI has released developer resources including best-practice guides, open-source example apps, and a chat-native UI library. An Apps SDK, currently in beta, allows developers to build experiences that integrate directly into conversational flows.

During the initial rollout, OpenAI’s monetisation is limited to external links directing users to developers’ own platforms. said it plans to explore additional revenue models over time as the app ecosystem matures.

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