needed. But if the AI is operating within our system, then we pay the cost of hosting and we’re paying the cost of meeting the request. If we’re doing more compute, that’s more expensive.” It is one of the many challenges TNA are grappling with: “There may be mechanisms for charging as part of that. What’s really interesting is how hosting companies are developing their offering to people like us. So, CloudFlare had a product that they launched late last year for charging bots. We don’t use this product, but basically it is like a pay per crawl product. There is another from AWS. So, you can see one of the things that’s hap- pening is that hosting providers are coming up with offerings for publishers, for not just controlling AI, but also for charging. And we have to pay attention to that because we’ve got the taxpayer to think about.”
Low friction
also going to cost X amount to provision 10,000 times the vol- ume of requests to our services – which it is not at the minute, but it could easily be – who should be paying for that? We have to assure value for money. We’re providing got services that are free for human beings to use. But we’re also now provisioning data for AI.
“We’re interested in exploring routes that will allow us to meet the need but that comes with an economic cost. It looks very small if it costs you 0.0001p to serve a webpage. But if that de- mand has moved into billions, suddenly that’s a lot of money. “This has been a mounting problem, steadily mounting. We are like a frog being gently boiled, over two or three years.” He explains how the order of magnitude rule changes the ‘thing’: “You’ve got your marginal costs that are traffic based, but then you’ve got a level of fixed costs with your infrastructure. But if the volume of traffic jumps up by an order of magnitude, espe- cially if it is computationally intensive, you need a fundamentally different infrastructure to meet that demand. So your provision- ing and your costs change quite a bit in that context. That’s the journey we’re on at the moment.
Let them in?
The MCP means a shift away from defence to cooperation. John says: “At the moment, we haven’t got anything like that in production. So we’re investigating this technology to understand it and see what difference it might make. We think it improves the quality of the results and it’s easier to provision, and that’s a win-win.”
But with every solution another problem arises. With the
MCP, one is how it might shift the burden of cost away from the AI providers. John explains that: “In the PDF case, where the AI deployed crawlers, the AI company paid for the computing power
“The underlying shift is the lower friction for a human being to initiate a process that creates thousands of requests. People don’t want to go to your website and search using your search. They want their AI tool to do it and their AI tool is going to issue a thousand requests. And there are more people using AI agents that issue thousands of searches and which are writing code and basically writing crawlers. And that’s just part of the change we’re going through. And then there are also just more people wanting to scrape content to train models.”
A similar drop in friction has happened in the pace of content creation which could read across to the volume of archives TNA may have to collect.
“In terms of inbound records we’ve not seen a big uplift. But if the transaction cost for producing a long document has reduced you can expect more content. And our anticipation would be – and I’ll put it no more strongly than this - that there will be more content because computers can now create content and that’s where we will need AI in information management to help us deal with the volume of content that’s being produced.”
Expect consequences
“If you look at like the history of technological development over the last 50,000 years, every invention comes with conse- quence. Our ancestors figured out how to rub two sticks together to create fire. Keeps you warm, enables you to cook, but it also dangerous. So consequences should be expected.” Asked if there is anything different or dangerous about the AI revolution, he said: “You have to be really careful about what you’re talking about. Different conversations have got mud- dled. Conversations about super-intelligence are very different from conversations about the large language models that we have today. I do not think, and this is my personal view, that we’re anywhere close to having a super-intelligence. I would be amazed to see that in my lifetime. But we are talking about extraordinary new technology. There are lots of things that we can now do that we couldn’t do before, including in archives. And there are some very real harms to protect and mitigate against. Navigating it all is now part of custodianship.” IP
‘‘ 26 INFORMATION PROFESSIONAL
If you look at like the history of technological development over the last 50,000 years, every inven- tion comes with consequence... So consequences should be expected. – John Sheridan
August-September 2026
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56
