Our July collection of articles and announcements to share this month.
Education
How Ofsted looks at AI during inspection and regulation – GOV.UK
Ofsted’s new guidance clarifies how they assess AI use during inspections and regulation in England. They’ve confirmed they don’t evaluate AI tools themselves but will examine their impact on learning outcomes, data protection and safeguarding. Inspectors won’t look for AI usage explicitly; they’ll only report on it if it significantly influences learners’ experiences.
Additionally, an Ofsted-led study investigated how 21 early-adopter schools and FE colleges across England are integrating generative AI tools to save teachers time on planning, resources, and admin. The report found that proactive AI use in education can ease workloads and personalise learning. However, many institutions are still developing curriculum-ready tools and lack long‑term evidence of real impact on student attainment.
Is UK higher education leading or lagging in global AI?
The UK sits fourth in global AI readiness, according to Oxford Insights. However, experts are warning that this lead is precarious. Despite growing interest in AI degrees, particularly among women, the UK’s educational response remains fragmented and underpowered. Professor Kamal Bechkoum highlights the urgent need for a cohesive national AI education strategy. Without such a framework, and with competitors like the US, China, and Finland advancing fast, the UK risks losing momentum in the global race to shape AI’s future.
State schools falling behind in new AI digital divide – The Sutton Trust
A new Sutton Trust report, shows private school teachers are more than twice as likely to receive formal AI training than those in state schools (45% v 21%), widening a digital divide. Private schools are three times more likely to have formal AI strategies (27% v 9%). Teachers in state schools report lower confidence, 24% are not confident using AI compared to 15% in private schools.
Assessment, and Academic Integrity
The Office of the Independent Adjudicator (OIA) has recently published several cases centred on AI‑related academic misconduct, highlighting recurring issues:
- In multiple cases, universities leaned heavily on reports from AI detectors, without sharing these reports or metadata with students. They often neglected students’ planning notes and draft versions; materials that could have shown how work was developed and countered AI‑use allegations.
- Two cases involved international students where detection tools were used as primary evidence without acknowledging potential bias against non-native English writing. In one, the OIA found the provider misrepresented a student’s viva admission, and in another, the student wasn’t given meaningful opportunity to respond to unseen evidence.
- An MBA student challenged the quality and origin of their dissertation feedback after an AI detector flagged portions of it. The OIA sided with the provider, noting feedback was produced via a standardised human-led marking process, supported by an external examiner report. They reaffirmed that AI detectors are not conclusive and human judgment and moderation should carry more weight.
Opinion
On the Sensibility of Cognitive Outsourcing
Derek Bruff pushes back on a recent MIT study suggesting that using ChatGPT to complete SAT-style writing prompts leads to reduced neural engagement and poorer recall. He argues the findings say more about the uninspiring nature of the assignments than about AI’s cognitive risks. According to Bruff, it’s no surprise that participants disengaged; these were artificial, low-stakes tasks with little relevance to their lives.
He also critiques the way such studies simplify writing into a one-dimensional activity, ignoring its complex stages where AI might play a more nuanced role. Bruff calls for more realistic research into how students use AI on meaningful, context-rich assignments. For him, the key isn’t banning AI but designing tasks where human thinking still matters.
Environment
Amazon’s carbon emissions jump as AI push tests company’s climate pledge
Amazon’s carbon footprint jumped 6% in 2024, rising to 68.25 million tonnes, primarily due to increased data centre construction to support AI and expanded logistics operations. This marks the first emission increase since 2021, presenting a growing challenge to its Climate Pledge goal of net-zero carbon by 2040, especially as AI demands surge
Research
‘Positive review only’: Researchers hide AI prompts in papers – Nikkei Asia
Researchers have revealed that 17 research preprints featured hidden prompts instructing AI tools to supply only positive feedback. These prompts, often embedded in white text or ultra-small fonts, appeared in papers from 14 institutions across eight countries. The journal Nature independently identified 18 similar cases, showing this isn’t an isolated incident. Critics argue that once AI becomes embedded in peer review practices, such prompt-tampering could spread, threatening academic trust unless publishers, institutions, and conferences enforce clearer policies.
Isambard-AI, the UK’s Most Powerful AI Supercomputer, Goes Live | NVIDIA Blog
The UK’s new AI supercomputer, Isambard-AI, is now up and running in Bristol. It’s the most powerful in the country and will support research in areas like climate, healthcare, and education. Isambard-AI was built in under two years and runs entirely on clean energy. It’s designed to be super-efficient and is already helping with NHS health research and developing AI tools tailored to the UK. This £225 million national project supports a wide range of research, from early dementia support to new materials science. Universities, researchers, and businesses can all apply to use it.
Public attitudes
This tool strips away anti-AI protections from digital art | MIT Technology Review
LightShed, a tool created by university researchers at Cambridge, Darmstadt, and Texas San Antonio, has been created to defeat established anti‑AI “poisoning” tools such as Glaze and Nightshade. These earlier methods added nearly imperceptible distortions to artwork that confuse AI training systems, either misidentifying style (Glaze) or subject matter (Nightshade). However, LightShed learns to identify and strip these perturbations by comparing “poisoned” and clean images, effectively restoring the art to a form suitable for AI ingestion.
Experimental results are concerning, LightShed detects protected images with 99.98% accuracy and removes protections reliably. Far from advocating for misuse, the developers argue LightShed is a wake-up call; revealing that today’s protections offer only temporary barriers and necessitate more durable solutions. The researchers plan to present their findings at the USENIX Security Symposium in August, urging collaboration to build more robust defences.
Vendor news
AI features for educators coming to Microsoft 365 Copilot | Microsoft Education Blog
Microsoft Education has outlined its latest AI-led tools designed to support teachers and students. Key highlights include the expansion of Microsoft 365 Copilot—now offering new reasoning agents like Researcher and Analyst, Copilot tuning for lesson customization, and deeper integrations with Learning Management Systems. They’re also launching Microsoft Learning Zone, a free AI-driven learning app (formerly Project Spark), enabling educators to create adaptive activities with content from partners like NASA and Kahoot. Copilot Chat is due to be made available to teen students in late July.
AWS is launching an AI agent marketplace next week with Anthropic as a partner | TechCrunch
Amazon Web Services (AWS) launched a dedicated AI Agent Marketplace on 15 July at its New York summit, offering a one-stop shop where AWS users can browse, purchase, and deploy autonomous agents. This is powered in part by Anthropic’s Claude models, with startups able to list their creations and AWS taking a cut of sales.
Introducing ChatGPT agent: bridging research and action | OpenAI
ChatGPT Agent combines ChatGPT’s conversational AI with deep research and automated browser navigation to tackle complex workflows end-to-end – from booking restaurants to generating spreadsheets. It’s available now to Pro, Plus, and Team users, with Enterprise and Education access coming later this summer.
Tips
Can We Teach our Moms to Spot Fake Ai Videos?
Finally, on a slightly more casual note than our usual articles, a YouTube group have challenged their Mums to distinguish between genuine and AI-manipulated videos. Over the course of the experiment, they dissect what visual clues give deepfakes away. It’s worth watching if you’re interested in how everyday viewers can better navigate the growing presence of AI-generated content and what signs to watch out for.
Find out more by visiting our Artificial Intelligence page to view publications and resources, join us for events and discover what AI has to offer through our range of interactive online demos.
For regular updates from the team sign up to our mailing list.
Get in touch with the team directly at AI@jisc.ac.uk