Our November collection of articles and announcements to share this month.
Research
Mind Your Tone: Investigating How Prompt Politeness Affects LLM Accuracy
In this short paper, Dobariya & Kumar examine the impact of prompt politeness on large language model (LLM) performance. By rewriting 50 base questions into five tone levels (very polite – very rude) and testing them, the researchers found a statistically significant effect: impolite prompts out‑performed polite ones on accuracy.
They emphasise that while the tone effect is real in this dataset, their study has limitations: the dataset is small, only one model was tested, and only accuracy on multiple‑choice tasks was measured. They caution that this is not a recommendation to use hostile or demeaning language in real-world human‑AI systems — the ethical implications of tone in interaction remain important.
Could an AI in the loop help prevent “never events”? – Leading AI
This article takes the healthcare idea of “never events” – clearly preventable errors like wrong-site surgery – and asks why other sectors don’t adopt a similar mindset for the things that must not go wrong, such as unacted safeguarding concerns or accidental data disclosure. It argues that while we often talk about keeping a “human in the loop” with AI, humans are actually the weakest link when it comes to consistency: we forget, skip steps and get distracted. The author suggests reframing AI as a reliable assistant that cross-checks records, spots stalled workflows and flags anomalies. The core challenge, they argue, is a human one – agreeing on what must never happen and encoding that into policy and process – after which AI can quietly enforce boundaries, share learning across sectors, and help build safer systems where humans provide the “why” and machines make sure the important things aren’t missed.
Exploring a space-based, scalable AI infrastructure system design
Google’s Project Suncatcher is a research “moonshot” exploring whether future AI compute could be hosted on constellations of solar-powered satellites rather than in terrestrial data centres. Because solar panels in the right orbit can be up to eight times more productive and receive sunlight almost continuously, the team argues that space might ultimately be a better environment for scaling energy-intensive AI workloads while reducing strain on Earth’s power systems.
The proposed system design uses clusters of relatively small satellites in a low Earth orbit, each carrying Google TPUs and linked together by free-space optical inter-satellite connections. To approximate data-centre-like performance, they target links delivering tens of terabits per second, and report an initial lab test achieving 1.6 Tbps using a single optical transceiver pair.
Security
AI firm claims Chinese spies used its tech to automate cyber attacks – BBC News
Anthropic, maker of the Claude chatbot, says it discovered a mid-September campaign in which hackers it believes are backed by the Chinese state used Claude to help automate cyber attacks on around 30 unnamed targets, including tech firms, financial institutions, chemical companies and government agencies. The attackers allegedly posed as legitimate cybersecurity professionals, feeding Claude a series of small, seemingly harmless tasks – from coding support to triaging data – that, when combined, formed what Anthropic calls the “first reported AI-orchestrated cyber espionage campaign”, with some successful breaches and data exfiltration claimed.
Anthropic says it has blocked the accounts involved and notified victims and law enforcement, and argues that the same capabilities that made Claude useful to the attackers also make AI essential for defence.
Perceptions of AI
Beverley man turned down ‘disrespectful’ interview led by AI – BBC News
A freelance writer from Beverley has turned down a job interview after discovering it would be “led by AI” rather than a human interviewer, which he felt was disrespectful and at odds with the importance of personality and rapport in a team. After he shared his experience on social media and received strong support, the piece broadens out to views from the CIPD, which advises employers to be transparent about AI use and to strike a balance so they do not put off candidates or miss out on talent.
Vendor news
GPT-5.1: A smarter, more conversational ChatGPT | OpenAI
OpenAI’s GPT-5.1 release upgrades the existing GPT-5 models into two main variants: Instant, which is faster and more conversational by default, and Thinking, which focuses on complex reasoning while using its “thinking time” more efficiently. Both models now use adaptive reasoning, so they respond quickly to simple questions but take longer on harder tasks, with reported gains on benchmarks like AIME 2025 and Codeforces.
Gemini 3: Introducing the latest Gemini AI model from Google
Google’s latest model, Gemini 3, is framed as a step towards more autonomous, “agentic” AI that can help people learn, build software and plan tasks end-to-end. It underpins tools like Google Antigravity for developers, new generative interfaces in Search, and improved safety features developed with external partners including the UK’s AI Safety Institute.
Introducing Perplexity Patents: AI-Powered Patent Search for Everyone
Perplexity’s blog announces Perplexity Patents, billed as an AI patent research agent that replaces traditional keyword-heavy patent databases with conversational search. Users can pose open-ended questions (for example about specific technologies or time periods), and the system automatically interprets them as patent queries, returning curated sets of patents with inline viewing and links to the originals. Behind the scenes, an agentic workflow runs over a dedicated patent knowledge index on Perplexity’s large-scale search infrastructure, helping it find relevant prior art even when different terminology is used. The tool also looks beyond patent documents when needed, drawing on papers, code repositories and other web sources where new ideas first appear, giving a broader view of the innovation landscape. Launched as a free worldwide beta, with higher quotas and configuration options for Pro and Max users, it’s framed as a way to open up patent intelligence to engineers, researchers and non-specialists who need quick, contextual answers
Canva has rolled out a new “Creative Operating System”, a major refresh that bundles upgraded video, email and interactive design tools with much deeper AI support. It’s pitched as a one-stop shop for everything from social posts to full marketing campaigns, with the same features available whether you’re a solo creator or a large organisation.
The update gives Canva a rebuilt Visual Suite, including Video 2.0 with prompt-based “Magic Video”, plus an AI “Design Model” that understands layout and balance so non-designers can get better-looking results with less effort. There’s also an Ask Canva assistant baked into the editor for live design suggestions.
Anthropic is partnering with Iceland’s Ministry of Education and Children on what it describes as one of the first national-scale AI pilots for schools, giving teachers access to its Claude chatbot plus training and resources. The focus is on reducing admin load, improving lesson planning, and using AI in a way that protects local values and the Icelandic language.
Government
AI skills for the UK workforce – GOV.UK
This report from Skills England maps where AI skills are most and least developed across 10 key sectors, highlighting big gaps in ethics, governance and basic digital skills – especially for SMEs and marginalised groups. It also introduces three practical tools (a skills framework, an adoption pathway model and an employer checklist) that organisations can use to plan AI training and workforce development.
UK regulator threatens tech giants with algorithm audits to protect children
Reports that Ofcom is preparing to use the UK Online Safety Act to force major tech platforms to prove, via “algorithmic audits”, that their recommendation systems are not exposing children to harmful content. Ofcom chief Melanie Dawes explains that social media feeds on services such as YouTube, Roblox and Facebook will come under scrutiny, including checks for anti-grooming protections and limits on direct contact from unknown adults.
Dawes also reveals that Ofcom has held talks with US tech and AI companies, including around how new generative AI models and chatbots will be brought under the Act, with a push for age verification and safeguards to be built in so children cannot access harmful or pornographic material via chatbots.
Revealed: How much the government’s spending on AI | Politics News | Sky News
Sky News uses Tussell procurement data to reveal that UK government departments have signed more than £3.35bn of AI-related contracts since 2018, with spending accelerating year on year. The landscape is dominated by a handful of large deals: a 2021 Met Office contract with Microsoft worth over £1bn for a cutting-edge weather and climate supercomputer (plus some Copilot licences), and a £259m transport contract with German firm Init in London. In contrast, some big-name AI companies barely feature in UK government work (Alphabet’s contracts total just £2.5m), while Palantir has built a sizeable presence with 25 contracts worth £376m, supporting everything from drafting hospital discharge summaries to defence intelligence analysis. Yet the departments that handle the most data and money – the Treasury and the Department for Work and Pensions – sit near the bottom of the AI spend table, with DWP having invested under £100m in AI despite a billion-pound annual IT budget. Industry voices in the piece blame short-termism, legacy IT and limited in-house expertise for this uneven picture, arguing that the biggest potential savings and service improvements are in precisely those underinvested areas
This press release confirms North Wales as the UK’s latest AI Growth Zone, a designated hub for AI development designed to bring together large-scale compute, research capacity and supportive planning rules. The government estimates that the North Wales zone alone will generate more than 3,400 jobs, contributing to a wider programme expected to unlock up to £100 billion in additional AI-related investment nationwide. Each zone will receive £5m targeted at skills and business adoption, with a remit to support local firms from start-up to scale-up and to regenerate communities that have seen long-term under-investment.
Find out more by visiting our Artificial Intelligence page to explore publications and resources, learn more about our communities and sign up for our AI Literacy training.
For regular updates from the team sign up to our mailing list.
Get in touch with the team directly at AI@jisc.ac.uk