Living in the grey
AI is not coming. It is here, and it is already working. Research suggests using generative AI saves the average worker roughly 2 hours per week.1 Professionals using it in controlled studies completed tasks 25% faster and produced work rated 40% higher in quality.2 In medicine, AI is doing things that were not possible before – reshaping how diseases are researched and compressing discovery timelines from years to months. At Triple Point, we are seeing the benefits directly, even in these early stages of our AI journey.
And yet, the same pace of development that makes AI so powerful is also outrunning our collective ability to govern it well. Being pro-AI and being clear-eyed about the risks are not in tension – they are, we think, the same position.
What responsible organisations need to stay on top of
We are getting out of practice at thinking for ourselves. A 2025 study of 666 people found a significant negative correlation between frequent AI use and critical thinking ability3 – most pronounced in younger users. The more people leant on AI, the less they engaged with what it produced. The researchers called it cognitive offloading. It is a quiet kind of vulnerability that compounds the more we ignore it. It also sharpens the moment access is disrupted. Uber, for instance, burned through its entire annual AI budget in four months.4 Dependency is part of the risk.
But cognitive offloading is only one dimension of the risk. In early 2024, a finance employee at the engineering firm Arup joined what looked like an ordinary video call with the company’s CFO and several colleagues. He was instructed to transfer $25 million. He did.5 Every other person on the call was a deepfake.
This story is particularly scary because it is no longer exotic. Deepfake fraud cost North American firms more than $200 million in the first three months of 2025.6 AI-generated phishing now achieves a 54% click-through rate, against 12% for traditional attempts.7 The technology to target any one of us by convincingly impersonating anyone whose voice or face is in the public domain is free and improving every month.
The question underneath the questions
The real question, we think, is not which AI risk is most pressing. It is how human intelligence and artificial intelligence are going to learn to coexist. Right now they don’t, comfortably. The work for any thoughtful organisation, in this moment, is to find a way of operating that lets both function well – without pretending the trade-offs are easy, or that the human cost of getting it wrong is small.
We have been here before. Nuclear technology. Genetic engineering. We discover what we can do, push it to an extreme, and then collectively pull it back and decide what we will and won’t accept.
Why this matters more in our sector
Adoption of AI is faster in financial services than almost any other sector – 81% of firms now use AI at some level, and 40% describe themselves as scaling or transforming with it, more than double the 20% of regulators at an equivalent stage.8 The gap between those using the technology and those setting the rules around it is wide. So is the gap between what AI can do and what most users fully understand about it.
That gap is where accountability gets uncomfortable. When an algorithm is involved in a decision and something goes wrong, the liability does not transfer to the algorithm. It sits with the person whose judgement was meant to sit behind it. The discipline of using AI well is, paradoxically, becoming a discipline of human judgement – knowing when to use it, how to interrogate what it produces, and that confident output is not the same as correct output. AI-generated content does not always reflect reality – and decisions have already been made on the basis of it. Trust, in this environment, has to be earned by the person putting their name to the work.
What we are doing – and what we are still working out
At Triple Point, AI is increasingly woven through how we work. We are developing an internal policy that sets out where it fits, what we use it for and what we don’t, and the principles underneath: appropriate equitable access across the firm, AI as a support for human judgement rather than a replacement for it, and ownership of any output sitting with the person whose name is on it. Some of what we have learned has come from getting things slightly wrong the first time – broad early experimentation that we have since brought back inside clearer perimeters, and having a more cautious look at some areas, such as AI-generated video, where the harm being done globally with the same technology has made us want to think carefully before contributing to it. We do not think this kind of course correction is a weakness. We think it is what responsible use actually looks like – at least at this stage, where the rules are still forming, the technology is still shifting, and the answers are still genuinely uncertain.
OK in the grey
Businesses operating in this space need to be okay being in the grey. The technology is developing faster than the regulation. The harms and the benefits are arriving at the same time. The work is to remain agile when it comes to responsibility – to keep asking the right questions even when we cannot fully answer them, and to remember that the discipline of asking those questions, again and again, is what protects the trust that earns us the right to be in the conversation at all.
We will keep sharing what we learn.
Want to talk through what responsible AI looks like in practice? Connect with us.
A note on how this post was written
This post was developed with the help of Claude, an AI assistant. Claude pulled together a research bank of current data on the AI risk landscape. Our team then debated it – using our own judgement to decide what was interesting, relevant, and worth saying – and Claude drafted the post based on that discussion. Every statistic was independently verified, and every editorial choice was made by us.
The content of this article is provided for information purposes only and does not constitute financial promotion or investment advice. The views expressed in this article are those of the authors and can change at any time. The content of this article is provided by Triple Point. In providing this document, Triple Point undertakes no obligation to procure any additional information or to update this document or to correct any inaccuracies.
Triple Point is the trading name for the Triple Point Group, which includes the following companies and associated entities: Triple Point Investment Management LLP registered in England & Wales no. OC321250, authorised and regulated by the Financial Conduct Authority no. 456597, Triple Point Administration LLP registered in England & Wales no. OC391352 and authorised and regulated by the Financial Conduct Authority no. 618187, and TP Nominees Limited registered in England & Wales no. 07839571, all of 1 King William Street, London, EC4N 7AF, UK.
1 Bick, Blandin & Deming, Federal Reserve Bank of St. Louis, 2025. Workers actively using AI saved an average of 2.1 hours per week (5.2% of work hours). https://s3.amazonaws.com/real.stlouisfed.org/wp/2024/2024-027.pdf
2 Harvard Business School / Boston Consulting Group, Navigating the Jagged Technological Frontier, 2023. hbs.edu.
3 Dr Michael Gerlich, SBS Swiss Business School, ‘AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking’, Societies, January 2025. doi.org/10.3390/soc15010006.
4 Fortune, ‘Uber burned through its entire 2026 AI budget in four months’, 26 May 2026. fortune.com.
5 CNN Business, May 2024; World Economic Forum (interview with Rob Greig, Arup CIO, February 2025). cnn.com | weforum.org.
6 Resemble AI, Q1 2025 Deepfake Incident Report. resemble.ai.
7 Microsoft Digital Defense Report 2025. microsoft.com.
8 2026 Global AI in Financial Services Report, Cambridge Centre for Alternative Finance, University of Cambridge. jbs.cam.ac.uk.