Two ideas have recently coalesced in my mind and have started a new train of thought.
The first idea is my relatively recent discovery of the term ‘AI slop’ (there is a Wikipedia article linked here). The ‘slop’ is low-quality writing and images generated by AI.
Given that AI has only just started generating, we might heroically guess that internet content is currently 95 per cent ‘good’ and 5 per cent slop. The numbers are actually unimportant – what matters is how the proportion is likely to change.
My hunch is that in 10 years we will be swimming in a sea of slop. We could let our imaginations run a little and consider new jobs and new businesses being formed to act as slop filters. Even so, I doubt that it will be super-hard to separate financial data from fake cat photos, even if we have to shift a mountain of fake to do so.
But we should introduce the second idea.
This one came from WEF’s Global Risks Report 2025. Unsurprisingly the biggest risk for 2025 is ‘state-based armed conflict’, followed by ‘extreme weather events’ and ‘geoeconomic confrontation’ (tariffs, anyone?).
In this context, I was slightly surprised by number four: ‘misinformation and disinformation’. Surprised because it ranked higher than 29 other possible risks, and further surprised when I read on and discovered that on a two-year view (ie today’s expectation for the most important risk in 2027), it is the top-ranked risk.
On a 10-year view it ranks fifth, but is the most significant non-environmental risk. A quick internet search (note to self, can I trust the results?) reveals that the University of Cambridge held a three-day disinformation summit in April 2025. It would appear that something important is going on here. Maybe there is a genuine threat to the integrity of financial data.
Before we merge these ideas it might be worth quickly distinguishing between mis- and dis-information. Both are incorrect information, but misinformation has no malicious intent. The objective of disinformation, on the other hand, is specifically to deceive.
So, WEF’s panel of 900 experts see misinformation and disinformation as one of the most significant risks over the next two to 10 years. And we have recently invented a fantastically efficient slop-generating machine that, in all likelihood, we intend to scale up massively.
Who do we expect to be the most adept users of this machine – the good guys or the bad guys? Is it reasonable to hope that the good guys will be able to keep us safe enough, as the anti-virus providers currently do? Or have the terms of the arms race shifted in favour of one side or the other?
One angle on this – perhaps the most important one – is ownership. In our hyper-connected information ecosystem, is it wise to rely on a handful of platforms, most of which are privately owned?
While we may be relaxed about the current ownership structures of the financial data providers, this may not remain the case, and we are relying on them to deal with the mis- and dis-information on our behalf.
I am not qualified to comment usefully on the disinformation angle, and so I am restricting my concern to the misinformation side, and the volume of slop that is likely to be generated. Will we be able to keep up in terms of distinguishing fact from fiction?
To give further context to my concern, in Thinking Ahead we have recently produced a fair bit of material on systemic risk and systems, usefully summarised in The simple, and enlightening, story of systemic risk.
In it we note that reality is too complex to understand, and so we build models – simplified versions of reality – in order to make our decisions. Implicit within this is the belief that the information we are using for our model building is an accurate or, at least, unbiased reflection of some aspect of reality.
Will this belief still hold in 10 years? Will we have the tools and skills to parse and filter accurate information from the inaccurate? Will the relationship between the market price and the unobservable true price of a security remain constant? Or might it change if the true price becomes too complicated for many to bother with?
And this is before we factor in the possibility that there is a growing band of actors with the intention to deceive us.
This is, of course, just conjecture. But all thinking ahead is the attempt to peer into the unknowable future given what is reasonable to infer from the present. If this is a feasible vision of the future then certain best-practice elements like governance and beliefs are probably worth a refresh.
In terms of new practice, or revised priority, I would suggest that the idea of data provenance is introduced and made a strategic priority. So each data point that plays a role in your decision making carries a flag showing its source, and perhaps a second flag that scores the trustworthiness of that source.
In this way, if you changed your mind about the trustworthiness of a source you could change the score rather than the data point, and allow your process to down-weight the influence of that data on your decision.
While we are at it, you could also play at pre-mortems – what if we discovered this data was unreliable, how would our decision change if we down-rated or disregarded it?
It is clear to me that the quantity of slop will balloon over the decade. It is not clear to me that this will necessarily be problematic – perhaps our existing tools, or new tools, will allow us to filter it out just fine.
Nevertheless, I can’t help thinking that making data provenance a strategic priority would be a sensible precautionary step. It is possible that this might also serve as a first line of defence against disinformation. However it is highly unlikely to help with the ownership issue we noted above.
Might we conclude that it is better, if more expensive, to generate the data ourselves (or via a trusted asset manager)?
Tim Hodgson is co-founder and head of research of the Thinking Ahead Institute at WTW, an innovation network of asset owners and asset managers committed to mobilising capital for a sustainable future.
