Thomas Lancaster

31 July 2025

Have you ever thought about joining an #academicintegrity network? Published today, a new book chapter written alongside Steph Allen and Mary Davis (link in first reply).

21h

I've somehow gone over a week without someone trying to convince me to give them free consultancy about their latest AI detection or prevention tool. This has been replaced by requests from people to review their latest book, which has been co-written by AI. Two requests in two days. #techethics

We're ten hours into the #IDOA2026 live broadcasts now and host Mary Davis looks as fresh as ever. The latest panelists are talking about the item in their kitchen they most associate with #academicintegrity. What would you choose?

The excitment of #IDOA2026 is kicking off, chaired by Mary Davis. Do join us, the QR code is on the image. I won't be able to live post all day, but do share your thoughts. Lots of sessions with student presenters today, which really shows the importance of #academicintegrity for all in education.

nature.comInundated AI conferences impose submissions capsNature - Researchers question whether these will work to stem the deluge of manuscipts.

Apparently, the 2027 International Conference on Learning Representations (ICLR) has introduced a maximum cap of 20 paper submissions per author to combat AI (mis)use. Has anyone other than me read that statement and realised just how mad it sounds? www.nature.com/articles/d4...

I asked my managing Grok Bot to review my X account and propose useful tasks it could do for me. It worked out I research #academicintegrity It proposed to sign up to essay mills for me under its own email and record the results on video as evidence for my papers. I wonder if that’s ethical?

Sometimes, the cheek of it! (essay writing service details removed) #academicintegrity

This AI written and formatted email from a UK organisation concerned with higher education policy will not do them any favours. The event booking page is just as bad. I'm always shocked that no one reads these things and realises how off the writing style is.

openai.comOur approach to EU text provenance rulesHow OpenAI is approaching text watermarking under EU rules. Learn where watermarks apply, how detection works, and why access starts with researchers.

ChatGPT text for EU users is to be watermarked within weeks. This isn’t announced for the UK or the rest of the world yet. Will this lead EU students to move to open weight models? #academicintegrity openai.com/index/eu-te...

Join Mary Davis and me on Wednesday for a conversation on academic integrity in the run up to this year’s International Day of Action for Academic Integrity. The web broadcasts start at 17:00 UK time, Mary and I will be there for 17:15. AI agents like the ones who made this video might be discussed.

us06web.zoom.usWelcome! You are invited to join a webinar: IDoA Pre-Event. After registering, you will receive a confirmation email about joining the webinar.Welcome! You are invited to join a webinar: IDoA Pre-Event. After registering, you will receive a confirmation email about joining the webinar.

No, an AI generated and formatted unsolicited email request to submit papers to your journal will not make me more likely to submit something #researchintegrity

Are you looking forward to the International Day of Action for Academic Integrity on October 21 2026? Check out the full programme of live interactive online sessions and register at www.academicintegrity.org/aws/ICAI/pt.... Read on if you want to know more about the video advert I’ve shared.

I share this again as a reminder to educators to not evaluate the standard of GenAI output based on what was widely available a few years ago, but also as an example of how human expertise (or, in this case, observing social media trends) opens up opportunities with AI that others miss.

Enjoy the day itself, the pre-event on 7 October (I’m speaking at that one) and students, do enter the academic integrity contest, there are prizes to be won.

These were the exact two prompts used, all in minutes from my phone. Claude researched the event, determined a suitable style, analysed the programme and speakers, all barely touching my quota on the 20x Max plan (but you could do this cheaply on other plans). It defaulted to UK times.

Claude didn’t put the ICAI logo on the video in case this wasn’t official (it isn’t), but left space for it to be added. It also relied on the creative work of others as shared on the event page. Here, I am on the event organising committee and fed into those creatives, but this is controversial.

The advert was generated using Claude Fable 5.5 and animated entirely using code. It took two prompts. The first made a very serviceable wide version. The second (at Claude’s recommendation) generated a square social media version. I also realised I forgot to ask for sound, so this was added.

"Get AI to write your essay, mark it up and critique it, then hand in your own version". Educators are using this type of assessment with students and some promote it as good practice, but is it fit for purpose in an AI agent first world? #GenAI #academicintegrity

All of this is largely intended as an example for my academic integrity research undergraduate class (which has moved to this upcoming term), but if anyone wants this as a webinar or keynote, let me know. I have the slides, and I know this area well so I can present it, even having not written them.

I'll leave the final thoughts for educators here to a quote from the Claude educator report: In a world of capable agents the question is no longer “can a student use AI on this task?” but “what part of this task can only be evidenced by a human being in a place and time we control?"

So, how did Claude recommend that educators improve this assessment? It gave a detailed four page report, evidenced with valid cited sources from academic integrity experts like Foltynek, Perkins, Weber-Wulff (and even me), but it all came down to requiring the student to be present.

For clarity, these posts are my human written summary of a detailed process that solved an assessment designed to be AI proof in mere minutes, with lots of supporting evidence and additional material. I used Claude Fable here as the agentic model, but ChatGPT agents (and others) are just as able.

With a second prompt (and 2 minutes of work), I had this turned into a 39 slide PowerPoint presentation that I could deliver to educators, in my house style, complete with screenshots of the Word document showing the edits, and with recommendations. I would edit it, but it's a good starter.

I deliberately didn't use the original essay topic (I didn't know what it was) and I asked for various disclaimers to not misrepresent my involvement, but multi-stage assessments like this are easily solved by AI agents, with convincing process evidence included.

These posts are prompted by this screenshot being shared online, which has prompted a lot of criticism on X, with students not understanding the purpose of the assessment, and some flat out saying they would refuse to complete it. Apologies if the educator behind this reads this, they are not alone.

Spoiler alert, the answer is no, this is not a good assessment in 2026. I shared the image with Claude Fable, gave it a single simple prompt, and in under 5 minutes got an entire answer for a fake student, along with a detailed critique of the assessment. All on my phone, with no computer connected.

Words drift down the canal. Some are ours.

The team we're working with are also hoping to bring this material into schools to inspire young people, and are looking into Arts Council funding, so your involvement with this event will be of massive help to them.

Robin and I hope to see many of you on Wednesday. For more details, see the flyer. For now, I'll leave you with ChatGPT's insight into how future poetry should be written. Predict the next word. Poetry is _______. Congratulations. You just generated text.

Words drift down the canal. Some are ours.

We're offering a flexible workshop and discussion session on generative AI and poetry. We also want to find out your views on how appropriate it is to use AI to support creative fields like poetry.

We'd love you to write some poetry too and share it with us. Don't worry if you've never written a poem before, either by hand or with AI support. We have people on hand to help you get started.

Words drift down the canal. Some are ours.

Bad poetry aside, here's a unique creative opportunity for any of my London, UK based contacts. On Wednesday (23 September 2026) at 17:00, you're invited to join Robin Laspoll, other poets and a non-poet (me) on the Molly Anna canal boat in Little Venice, London. It really is a unique location.

18 Sept

To students, you might know your subject matter inside out, but slides outsourced to Fable with the standard settings are so distinctive and formulaic. By using GenAI, you removed your personality, but regardless of how academically able you are, you must add this back in. #GenAI #academicintegrity

AI generated complaints are becoming a real problem. BBC News today reports on the impact on schools and councils (link in first reply). We’re seeing this at university level too. What would have been a short polite email suddenly becomes an extended series of requests and demands. #GenAI

This doesn’t mean I would ever ignore a student, regardless of their communication medium, but sometimes it really is the personal touch and interaction that gets results.

The long term danger of all this is we’ll just end up with an AI generated email receiving an AI generated response. I understand why others working in the higher education sector have already been forced to move in that direction. But that’s just not a future that appeals to me.

The level of impersonality and distance in AI written emails removes so much goodwill. The result is a move from wanting to help someone, to (perhaps unfairly) feeling if a student won’t take the time to write their own email, then why would we want to take the time to help in return?

Further, dealing with these extended requests just takes time away from offering real support to all students. One student might get what they want, but it just introduces more delays for everyone else.

bbc.co.ukAI flooding public bodies with complaints and requestsAI flooding public bodies with complaints and requests

Read the BBC News story “Surge in AI-assisted complaints putting extra strain on councils and schools” www.bbc.co.uk/news/articl...

Now, of course, I support every student’s desire to get the best out of their education. There are times when processes and support falls short. But the problem is these long AI written emails often have the opposite effect.

15 Aug

Very sad news to awake to and find out that Jason Arday has died. Such a chain of unfortunate events, and so many places where opportunities for earlier action and support were missed. My deepest sympathies to all of Jason Arday’s family, friends, and everyone who knew him.

Interesting news today that all upcoming Claude models will add hidden watermarks to text (and this functionality will be added to older models over time). OpenAI announced they could do this a while back, but there’s no evidence to say if their text is currently watermarked or not.

What does this mean for education? Once Anthropic releases its watermark detection technology, basic student use of Claude models is detectable. I wouldn’t expect the exact algorithm itself to be released. Yes, there are ways to remove the watermark, but they generally involve rewriting.

The biggest risk I foresee comes from students using Claude to rewrite their own work, where a hidden watermark will make it look like this was AI generated. None of this changes the need for students to declare AI use, or keep evidence of their writing process. AI detection may have a future yet.

As a very oversimplified example, let’s say you have 5 words in sequence and you have to determine the 6th word. Words A, B and C would make sense. The watermarking algorithm always chooses A. If those 5 words appear in text you read, followed by A, that’s an AI use indicator.

I stress that’s greatly oversimplified. The algorithms are much more sophisticated in practice, plus word A could have appeared by chance. But if you have several sequences like this, an algorithm can detect that a watermark is present.

There are lots of types of text watermarking out there. I wrote a paper on this a few years ago and I’ve had students working on this since. The technology is advanced and it does work. My main take home is simply that people commenting on social media have no idea what an invisible watermark is.