Our April collection of articles and announcements to share this month.
Education
AI literacy is not enough – universities must teach through disciplinary standards
Annika Bautz argues that generic “AI literacy” modules don’t prepare students for AI-mediated work. Drawing on the University of Surrey’s plans to embed discipline-specific AI teaching across all programmes from September, she explains why each subject must define its own standards of evidence, judgement, and risk if graduates are to develop genuine professional expertise rather than fluency without competence.
Accessibility
On AI glasses and wearable AI in assessment
This paper introduces the concept of “dual transparency” to argue that consumer AI glasses, which can read exam papers and display answers invisibly, undermine the conditions on which invigilated exams and orals depend. The authors warn that doubling down on physical exclusion risks pushing institutions into intrusive bodily surveillance rather than restoring assessment security.
Academic Integrity
AI in Assessment: What Ofqual Is Really Signalling
This post argues that Ofqual’s stance on AI in marking is less a restriction and more a call for intentional system design, anchored in validity, fairness, and trust. The piece pushes back on superficial “human-in-the-loop” thinking, asking sharper questions about where human judgement actually sits and how AI-supported decisions can be explained, challenged, and trusted.
Security
What is Claude Mythos and what risks does it pose?
A short BBC explainer on Anthropic’s Claude Mythos, the model the company says outperforms humans at hacking tasks and has only released to a small group of partners. It captures why finance ministers and central banks are paying attention, while echoing the NCSC view that the priority remains getting cyber-security basics right.
Project Glasswing: Securing critical software for the AI era
Anthropic has launched Project Glasswing, a coalition with major tech firms and others to use its unreleased Claude Mythos Preview model to find and fix vulnerabilities in critical software. The post details thousands of zero-days already uncovered, including 27- and 16-year-old flaws in OpenBSD and FFmpeg, and frames the initiative as an urgent push to give defenders the upper hand before similar capabilities spread.
Anthropic investigating claim of unauthorised access to Mythos AI tool
NCSC head Richard Horne told the CyberUK conference that frontier AI could strengthen cyber security if organisations get the basics right, even as Anthropic investigated reported unauthorised access to its powerful “Claude Mythos” cyber model. The piece highlights the tension between AI’s defensive promise and the risk of advanced capabilities leaking outside intended controls.
Perceptions of AI
The tensions of AI shouldn’t come as a surprise in a system wired for speed and output
Jo Irving-Walton reframes generative AI in higher education as less of a governance issue and more of a “dopamine problem”, offering cognitive relief in a sector already wired for speed and output. The piece argues you can’t out-policy a reward loop and that universities need to rethink what they value, not just tighten the rules.
Behind the Curtain: The kids aren’t AI-right
New polling shows Gen Z’s excitement about AI has fallen sharply while anger has risen, with most college students saying their schools either discourage or ban AI use. With entry-level roles, the very jobs that build judgement, the most exposed to automation, the piece warns that the generation best placed to thrive with AI is being left least equipped for it.
Vendor news
Introducing ChatGPT Images 2.0
OpenAI has launched ChatGPT Images 2.0, billed as a step change in detailed instruction following, accurate text rendering, and composition, with results said to feel “intentionally designed” rather than obviously AI-generated. It’s also OpenAI’s first image model with thinking capabilities, allowing it to search the web, generate multiple variants from a single prompt, and check its own outputs.
Copilot’s agentic capabilities in Word, Excel, and PowerPoint are generally available
Microsoft has made Copilot’s agentic features generally available across Word, Excel, and PowerPoint, allowing it to take multi-step actions directly inside documents, spreadsheets, and decks rather than just suggesting edits. Microsoft frames it as a shift from passive assistant to true collaborator, with a continued emphasis on user review and control over changes.
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