Panel Discussion on The Role of AI in Financial Services

Finance For The Future
12 followers
2 Video Views·Dec 31, 2025
Thank you, Rafael, for introducing the discussion.
Very interesting tests and pilots undergoing at the hubs.
So this panel is about AI in finance.
We know AI has been dominating the news and the agenda
and we have headlines almost on a daily basis, breaking news.
We have very mixed reactions going to the AI world
extremes at times from glorification
of the potential benefits to a doomed scenario
about the end of human primacy and overtaken by robots.
So it's a very, very interesting debate.
Today we have four amazing speakers with us
to help us better understand the impact
on the finance industry in particular.
To introduce the speakers, from my right hand side,
we have Tom Maton, who's Director of FinTech
at the Bank of England.
Then we have Svetlana Pankova, MEP.
We have Jonathan May with us, CEO of Lloyds Europe
and Claude Marx, Director General of the CSSF.
Thank you very much for being here with us
and without further ado, as I know everyone wants to listen
what you have to say, let's start with the current state
of play of AI in finance.
And starting with perhaps from the right hand side,
the question addressed to everyone.
What is the current state of play in finance specifically?
What are the use cases that you find are most in traction?
Are these material to the firms and the sector overall?
And where do you expect to see most of the applications
of advanced generative AI models moving from pilots
to actual live production?
Tom, please.
So thank you for the invite to speak to you today.
So 75% of UK firms are using AI.
That's in a joint Bank of England SCA survey.
They report that 62% of those use cases
are what they deem to be low materiality
from a risk and criticality to their business perspective.
But interestingly, 16% of them are deemed
to be high materiality.
Now obviously they can all define that in the way they wish
but that gives you some kind of sense of the baseline.
And when we've asked them about specific use cases,
by and large, it's the sorts of things
you would expect to see.
So things like chatbots for customer services,
AML, CTF, internal processes, HR, et cetera.
But some of the sectoral differences are quite interesting.
So for example, and this is all public, by the way,
you can read it on our websites.
We publish all of this information.
For example, we see quite a lot of use
including in quite advanced deployments
in consumer credit, for example.
We see less in financial market infrastructures.
Banks a little bit further ahead of insurers
but not by loads.
So quite interesting as you sort of break some of this apart
but overall AI has arrived, will continue to grow.
Relatively speaking, the majority of deployments
are low materiality but we can see some high materiality
and that will, again, will grow over time, I think.
There's two trends which I do think
are really interesting there
which have come out loud and clear in the survey.
One is just the start of evidence of growing complexity.
So some of the models that people are reporting using
are probably on the more sophisticated end
and 17% of those models which were reported
is being used for foundation models.
So again, on the very complex end.
Clearly looking for automated decision-making
is something we're very interested in.
Firms reported that 55% of models
involve some level of automated decision-making
but actually, interestingly, only 2% of those models
were fully autonomous.
So no human in the loop.
We'll be watching that one very closely.
We repeat this survey every couple of years
to give ourselves a time series.
So evidence of growing complexity,
nothing to worry about right now, I don't think.
But something we need to keep a very close eye on.
The second kind of key theme
which is something that Petra spoke about
and Rafael spoke about as well
is the growing emphasis on third parties
and the criticality of those third parties.
In the most recent survey,
a third of the AI deployments
used external vendors models.
And within that, we see a really quite sharp
increase in concentration
over the time we've done this survey
but a very sharp increase in concentration
in the most recent iteration
towards the top three providers.
So we asked them to split it apart,
cloud, model, and data.
And you can see the concentration in the top three
was 73% of cloud providing,
44% of model and 33% of data.
Going up quite a lot.
Some very clear regulation policy implications there.
And then the very final thing I'll say is
we saw a really interesting kind of dissonance
that we asked firms to talk about
their level of kind of comfort with AI
and the technical dimensions of AI,
particularly in their governing bodies.
Around 40% of them said they felt they had a good grip
on the technical side of artificial intelligence.
Doesn't surprise me at all.
I don't think it's necessarily reasonable
to expect a governing body to understand all of this.
The thing that's quite interesting though
is when we asked them to rank
the risks they were worried about,
all of the risks they were most worried about
were exogenous.
So it's quite interesting.
So there were things like privacy, cyber.
It's quite interesting, I think,
to have somebody say, I don't really understand this,
but by the way, the risks I'm most worried about
are things that somebody else can do to me
or environmental risks in some way.
I'm not at all surprised that cyber, privacy, et cetera,
were the sort of highest profile risks.
But that sort of dissonance is an interesting one, I think.
So I'll stop there.
Thank you, Tom.
Perfect, so I will disturb the chain now
and give a bit of a regulatory perspective here.
So I think we all know the benefits
and the opportunities that AI offers us.
We've been speaking about this since the morning,
more or less, in all the interventions so far.
And AI is being already implemented
in a lot of areas and aspects
of the way the industry functions,
and especially the financial sector.
So if you allow me, I will just very briefly focus
on some of the challenges that remain.
Because further on in the discussion,
I would like to build upon the reasoning
why we are still focusing on making a strong
regulatory framework to ensure a level playing field,
which is fair for financial services,
for the consumers, and for the industry of the EU.
So I will focus on five areas
which we still find challenging,
or like we still find that contain certain amount of risk
when we're speaking about the implementation of AI
in the financial service sector.
So the first one is data security and privacy.
This would remain our core priority
when we're building the whole regulatory framework,
when we are focusing, as we do now,
on the AI Act implementation.
Because AI would rely, you hear me well, right?
Perfect.
AI will rely on large amounts of very sensitive data.
And our responsibility is to make sure
that this data is safeguarded.
Because when we speak about private data,
we know how a breach of this
can actually destroy the trust in the whole system.
Which basically leads me to the second challenge.
This is consumer trust.
We speak a lot about AI applications
in all areas of our life at the moment.
But allow me to have this judgment
that when we have the financial service sector,
there is a higher standard expectation for the consumers
when we speak about the trust level there.
So they need to have faith in the financial institutions
and the instruments that have been used.
And here it's another aspect also,
which applies to this part, this is the literacy.
We are facing the challenge sometimes
that the financial literacy is not at the high level
as we want it from the European customers' perspective.
And that's why when we're speaking about customer trust
in the new technologies and AI application
in the financial sector, we have to be aware
that the standard and the expectation
from the receiver side is much higher
than any other area where we see the application of AI.
Third is, of course, transparency and accountability.
You see that they're all going in a chain
and they're more or less connected.
But there must be like a very well-communicated
and well-explained algorithms
used by the financial institutions,
which ensure that there is transparent
and accountable way of tracking the data usage
and the results that have been generated by AI technologies.
The aspects and fields in the financial sector.
And the last two, I'll be very brief, fairness.
What we see when we look at different data and studies
is like the monitoring for any discriminatory practices
and patterns when we're speaking about AI application
is starting to be quite important
also from the regulatory perspective
because we wanna make sure that this is avoided.
And last but not least,
we have to acknowledge the human factor as well
because AI is here to supplement
rather than replace completely the human expertise,
especially when we're speaking about judgment,
experience, then giving a proper financial advice
to both retail customers
and of course the professional investors.
So we need to make that differentiation
and ensure that we are working towards
striking the right balance into that mix.
I will stop here now.
And I'm sure we're gonna touch up
on some of the topics further on.
Exactly, we'll come back to this.
So data privacy, consumer trust, including literacy,
transparency and accountability and fairness
and keeping the human in the loop
if I can summarize your intervention.
Now, going to Jonathan,
going back to the state of play
from the industry standpoint,
could you give us your view for the insurance sector
over and above what Petra introduced
in the introductory remarks?
And also would be very interesting to hear from you
your perspective on this potential risk of dependence
on a small number of third party providers
for AI related services that Tom mentioned.
Sure, thank you.
So yes, as Petra mentioned,
it's still quite early stages,
the level of AI use within the industry.
I think reference was made to
the sort of back office functions
that we like improving the efficiency,
improving the claims handling experience for the consumer
and also compliance as well checking
because it is a heavily regulated industry.
There's lots of compliance checks that we need to do.
So using it there is definitely coming through.
But even there, I think, I echo the points.
We need to be very mindful
that this is to the benefit of the consumer.
So is it improving the experience to the consumer,
the use of AI in the back office,
in the claims experience.
So if you've just lost your house to a natural disaster,
do you want to speak to a bot
or do you want to speak to a human being?
So I think we always have to have the value
and the experience for the consumer at the front of mind.
We are also seeing, again, new risks coming out,
excuse me, for which we need to provide solutions for
as an industry.
So whether that is sort of AI models themselves
and providing insurance for the providers
of those models, et cetera,
or other things such as self-driving cars, et cetera.
So the technology itself is then creating new,
a demand for new products from the industry.
On the front office, as it was said earlier,
this is still very, very early stages.
What we are seeing is algorithmic underwriting
where you've got a lead underwriter's decision
and you've got sort of follow capacity
decided on a algorithmic basis
to more quickly get that capacity in place
to meet the demands of the customer.
I think it's fair to say that in different parts
of the sector, the use of AI is greater.
So Lloyd's is predominantly wholesale,
wholesale insurance market
where we're bringing different parts of the ecosystem,
the policyholders, the brokers, the carriers together
and we act a little bit like a quasi regulators.
We're doing largely wholesale insurance
and insurance like motor, obviously AI
is probably much, much more advanced,
the high volume classes of the business.
But we can see the potential for augmented underwriting.
So a lot of the tasks that an underwriter has to do,
we can use AI to augment their underwriting
so that the underwriter is focusing
on where you need that human position on the complex risks
and is then supported by the technology
to enable them to make those judgment tools
which should make products for existing risks
much more accessible as well.
Thank you.
Claude, moving on to you,
it would be interesting to also hear from you
how you address the current data gap
in terms of the deployment of AI in finance.
We have a lot of evidence that is in a sense anecdotal
in many jurisdictions.
The data gap in this area is one of the policy priorities
also at the G20 level with a deliverable
coming to the South African presidency later this year.
How do you see this issue of data gap
and also what's your experience
with the deployment in Luxembourg?
Yes, good morning and thanks for having me here.
In terms of the, let me start
with the last bit of your question.
So what's the experience in the Luxembourg Finance Centre?
We have just surveyed some 500 financial services providers
which is a very large group of people that we supervise
and more than 80% of people surveyed
have participated in the survey.
And what we have seen is echoing actually
what has already been said by Bruce and others this morning
is that we are seeing slow adoption.
Currently only 28% of people surveyed
have actually in production or are testing AI systems
and 22% want to do so in the next 12 months.
So that gets us to 50% which is significantly lower
from the experience that Tom has at the Bank of England.
But I think it's largely reflecting
what we see in continental Europe
if you look across all the industries.
So be it asset managers, be it banks, be it fund managers.
And then in terms of what is being used,
it's largely back office efficiency type of usage,
use cases that we see and very low percentage
in products and client facing systems.
But it's true that we are really facing a data gap here.
There are many generalist surveys out there
by big four and other consultants
that give a very high level oversight
of what is being used.
But we are lacking data on the effectiveness
of systems being used on the risks
that these systems generate
and the risk mitigation measures put in place
by financial service providers.
We are also lacking data on the long-term impact.
And so we have a regulatory blind spot here
and that means that it is very difficult
to have evidence-based policy decisions
at this point in time.
Now, what we could do to close this data gap,
I think we could have standardized AI reporting
and transparency from the market.
For instance, what the number of AI systems
are used, the type of systems being used,
the data sources, key performance metrics, and so on.
We should also have independent research
and public data repositories
where we overcome the confidentiality issues
in that anonymized data is populating these repositories.
And then we need cross-sector collaboration
and data sharing amongst regulators,
market, academia, and tech firms.
I think these are a few things
that we should be focusing on in a collaborative effort
to help us to overcome that data gap.
Thanks, Claude.
That's very, very interesting indeed.
So going back to regulation and back to Svetlana,
you've given us a little bit of an overview
of the risk and the risk mitigation
when it comes to the use of AI in finance.
Now, looking more at the bright side of the story,
what is the expected impact of the AI Act
on the European financial sector more broadly
from a positive standpoint?
How can AI, for example, help integration
of the EU single market?
Can it also help with the capital markets union project?
What is your view on that?
I also tried to present some opportunities
together with the challenges,
trying to justify what we do in the regulatory framework
because, I mean, we have to be very fair and honest
about the possible risks and the problems
we might be facing.
That's the only way to actually create
a stable framework from that perspective.
So a very important question,
like the capital markets union
and like the whole functioning
in the financial sector in the EU
is one of the key topics currently all over Europe
in light with the competitiveness aspects,
which was discussed from the first talk this morning already
it was one of the key topics.
So definitely the capital markets union
is one of our priorities and finally making it work.
So AI is definitely a part of it
and I would just actually focus here
on three key points which would help us build
that capital market union working
because the first problem is the fact
that we don't have harmonized financial services
in the regulatory framework across the EU.
That's the real and the pragmatic reason
for the postponement of creating
the well-functioning capital markets union
and AI could actually be very helpful
in trying to make sure that we find
the smoother cross-border cooperation
and especially ensure that the regulatory framework
could be harmonized in the most efficient way.
So the biggest challenge in the most obvious application
in AI is in that aspect.
Of course, step two is to make sure
that the institutional investors
and the professional investors,
they see the value added in the cross-border transactions
and actually try to shift a bit
the level of risk averseness
because that's one of the biggest challenges
when we're speaking about the functioning capital markets
in the EU and last but not least
is the smooth liquidity and capital allocation
where AI also plays a key role.
So first, regulatory harmonization.
Second, the professional and institutional investors
should actually start moving capital
and investing it in a more efficient way within the EU
and last but not least, once all that is ensured,
the capital allocation would boost all the participants
in the financial sector to be part of this.
So there are many more things to say here
but I think those are the three quite obvious steps
and a very visible role of the AI in that field.
Jonathan, Svetlana touched on competitiveness
and the potential impact the AI Act is having
on the competitiveness of the European financial sector.
How do you see from the industry standpoint
the consequences of this regulation for competitiveness?
Also taking into account particularly pertinent
the potential controls for general purpose systems
such as LLMs and other generative AI models.
Sure, thank you.
So I think we all agree we want to stay competitive
and we want to be innovative in terms of having regulation
that is pro-innovative if you like
but obviously I think one of the ways
that the EU AI Act can help with the innovation
is that it does provide regulatory clarity
and I think particularly for financial services
that having ambiguity is in no one's interest.
It's not in my sector's interest
because the risk of things going wrong can be quite large
and therefore what you want to avoid
is people just doing a wait and see kind of stack.
So having that clarity is beneficial definitely
and obviously this is one of the first pieces
of legislation out on AI globally
so I think that's to be welcomed.
But as I said, it's important that we do remain competitive
and we also have to recognize
that this is a massive and rapidly evolving area
and the risks in this area are evolving
so we need to keep that dialogue going.
And we also need to, between supervisors
who are perfectly placed to see the opportunities
and the challenges across the different layers
within the financial services ecosystem
and support coordinator, you can only,
you need to have those sort of grown up conversations.
This is what we're seeing, this is the risks
that we're seeing, this is how it's changing
how we want to operate and having those conversations,
sharing the data, sharing the information
I think is absolutely crucial
because the risks will emerge, may be the wrong word
but greater awareness of the risks
as the technology gets more and more developed
I think is key.
So I think this is great, the app's great
because it removes a lot of the ambiguity
that was creating the sort of paralysis, if you like
and now you've got that certainty
so it does help AI service providers.
A lot of the discussions this morning were touching on data
whether it was Vida or Rafael's discussion
about specific use cases and pilots.
So I'd like to hear from Svetalina
your view about the interplay between the AI Act
and other EU regulation such as the Data Act
but also GDPR or even the Data Governance Act.
So the short answer is the AI Act is in harmony
with everything else that we have set
in terms of regulations over the ones you've listed.
Please allow me to explain also the infrastructure
we're trying to build with all those legislations.
I know it's still the first panel in the morning
but recall back some of the challenges I've presented
and now I'm gonna try to explain how are those resolved
with the legislative infrastructure we are trying to create.
Actually we have created and now it's about time
to see how the implementation will go.
So if we combine the GDPR with the AI Act
then we get the answer to the privacy
and the security of the personal data.
So making those two work together
would resolve one of those challenges.
And when I say resolve, we're not speaking about
like some magic solution which seizes the problem
to exist in the future but it is offering the ways
to solve it for the time being
and for the long term of course.
Data Act plus the AI Act,
it is all about the quality of data.
So basically it's ensuring also some of the ethical standard
but it's about the quality, ensuring the quality of data
and again safeguarding it.
You've asked also about the Governance Act.
So if we combine that with the AI Act,
again we're speaking about sharing of data
in line with all of our high standards.
So we have the privacy, you have the quality
and now you have to share it
because it has an economic value
and that's the only way to make it actually work.
And last but not least, you've mentioned FIDA.
If we combine it with the AI Act,
it touches upon one of the last challenges
which I've outlined before.
It's about the fairness standard
especially in the high risk areas.
So this is a map which basically shows
that understanding all the challenges,
we've tried to create an infrastructure
which gives us the tools to be able to overcome,
resolve or address them in a timely manner.
But it was already touched upon by all the speakers
around me actually and by the previous interventions as well
that we need regulatory clarity and predictability.
But now it's also the time
when we have to offer some flexibility
given that the technological advancement is very quick
and if we wanna actually speak about true competitiveness,
we have to be able to look around us
and see what's happening.
So I'm very confident that we have created
the right framework and infrastructure.
Now it's about time to implement it
but with a very common sense of having the flexibility
which makes it work given the fast,
rapid investing technologies.
Since we're good with time,
I'd like to tease Tom with the question of competitiveness
that keeps coming up.
From your standpoint in the UK,
we are reading everything about the explicit criticism
to the EU about the over-regulation
potentially hindering innovation
and not allowing the beneficial aspects to come up.
We also understand that regulatory clarity
and risk awareness as Jonathan mentioned
and this reduced uncertainty helps promote innovation.
How do you see all that from the competitiveness angle
from where you sit in the UK?
So maybe to completely dodge your question
and ignore the competitiveness part of it,
let's think about growth rather than competitiveness
So look, we want growth, we want innovation,
we want to unlock the productivity puzzle.
We have a bunch of, it's not just in the UK,
this is probably true across much of the Western world.
We also have a bunch of structural challenges
which do lean on growth, aging population, et cetera.
So that we need to use AI across our economies
is kind of totally uncontroversial.
It would only be sustainable though
if this is used safely.
And the question is how do you give people
that right balance between users,
potential subjects of AI processing,
knowing that this has been safe,
done in a manner that accords with their societal norms
and their expectations,
but also isn't the wrong fit for something
which the rate of progress here
is measured in days at the moment.
So it's not a natural fit for very detailed regulation.
I think when we apply this in the UK context
and in financial services,
the first thing I say is obviously,
we contribute to broader UK government efforts
and the UK government has collaborated
with colleagues around the world,
including European Union.
So we contribute to broader UK government efforts.
Obviously that is focused on the most high risk deployments.
Financial services isn't currently one of those
in the view of the UK government.
When we drop that down then into finance,
where we work very closely
with the Financial Conduct Authority,
we jointly had an artificial intelligence
public private forum.
We asked them to write a report for us
on barriers to deployment and use of AI.
We published that report.
They didn't really identify regulation.
They basically said regulation wasn't a barrier.
That really poses a bit of a puzzle
because it's very clear where they see the limitations.
It's access to compute, access to skills, access to data.
It's these big sort of infrastructure
and structural questions.
They would like some more regulatory clarity at the edge.
The areas they identified for regulatory clarity
were a bit more guidance around expectations
of senior management and those in governance functions
and a little bit more around things
like model risk management policies.
Realistically, providing more guidance on those,
which is something we're very kind of willing to do,
probably isn't going to resolve
the slightly more fundamental point
that quite a lot of firms say to us,
look, there isn't a specific regulatory barrier.
There's not a thing where we say,
if you just change that rule, this would be much better.
The thing they're worried about
is this broader question about uncertainty.
And when something is moving so quickly,
when what you're dealing with,
I think particularly in the consumer space,
starts to touch on social norms, your social values,
broad concepts like privacy, which sort of transcend finance,
you get into the sort of dynamics
that Petra was talking about, right?
And there, you may find that regulatory certainty
is fleeting because society is moving
around these technological developments.
So I think there is a big question there
about how can we best make sure
that we're not stifling the innovation,
we're staying on top of the progress,
we're acting early where we see actionable risks,
but also not too early.
And I think for us, we're very happy to issue guidance,
but really the starting point for us, I think, in the UK
is to deepen the conversation,
much in the way that Bruce talked about this morning.
So we've launched an artificial intelligence consortium
for academics, for technology experts,
for industry to come and have a conversation with us
in the Financial Conduct Authority
about use of AI in financial services,
and for us to really start to try and get into
some of the knotty technical issues.
So for example, how do we think about things
like algorithmic collusion?
How do we think about shared methodologies,
which may be hidden in models
and which lead to prosaic locality and herding?
And I think we've got to be humble as well
about what this would involve.
So in the financial stability domain,
which is kind of where I work,
we're very used to following objective,
largely source of output type measures,
asset prices, economic variables.
We don't look at kind of shared assumptions
going into models.
That's not something we generally do.
I mean, probably the closest we get is approval
of internal capital models.
That's light years away from dynamic,
constantly evolving models like artificial intelligence.
So I think starting to deepen this conversation
and think about how we can be agile
is going to be really key to unlocking the potential here.
Thanks, Tom.
You mentioned that based on your survey results,
that regulation does not seem to be
one of the major impediments
to the wider use of AI in finance.
So I'd like to turn to Jonathan and also Claude
for their views on that.
Yes, so one of the findings that Tom mentioned
was that regulation does not seem to be an impediment
to the wider use of AI by the financial sector.
Is this a view that you share from your day-to-day?
So today, no, but the regulation is obviously
just coming out, really, in all the jurisdictions.
So obviously, we talked earlier about the EU AI Act,
which does give certainty, but equally,
I do think there has been a degree of wait-and-see
to see what happens with some of the regulations.
And therefore, that's maybe impacted
to what extent AI can be utilized as well.
So I don't think it has been to date.
Obviously, as the regulation evolves,
it's important that everyone's sharing information
about the risks and having that open wire
to continue to get, because you need the right regulation
in place to manage the risks.
Now, whereas that's various different risks that exist,
and that it is proportionate while still creating
that sort of pro-innovative competitive environment
that we want, but you have to have the right regulation.
Yes, I think that the AI Act is not harming
deployment of AI, and it's a generalist regulation
that classifies essentially AI systems
into four risk categories.
There's no system that is forbidden in the European Union
under this AI Act that would negatively impact finance here.
We have a couple of high-risk systems
when it comes to credit scoring or in insurance.
Indeed, we have a couple of applications
that would be considered high risk,
but that's very manageable.
And then we have various sorts of regulations
that were already touched upon that also have an impact
on the deployment of AI.
We have market manipulation regulation.
We have mis-selling.
We have cybersecurity, DORA, data protection, privacy.
Probably the GDPR and data protection regulation
had the biggest impact on deployment of certain AI systems,
especially when it comes to using widely used
US-based generalist generative AI systems,
like ChatGPT or Gemini or Copilot.
But there are solutions that have been deployed
by the vendors to overcome those data protection
and privacy issues through local versions of those systems,
through air-gapped servers,
sovereign cloud systems and the like,
where all these systems are made available,
but in EU servers, not to fall foul of privacy
and data protection regulations.
So in summary, I think that the current state of play
is that the various pieces of regulation that impacts AI
did not have a harmful impact on the deployment of AI.
So Jonathan mentioned that this is an ongoing process.
So the next part of our discussion is about looking forward.
And I would like to follow up, Claude,
on what you just said.
Going forward, do you think there will be a need
to refine the rules?
Do you expect any gaps to be identified
and any further rules to be needed?
And to note also that we have run OECD analysis
late last year on the regulation of AI in finance
in the 38 member countries of the OECD.
And the result was that the vast majority of jurisdictions
in OECD membership, which goes from US, Canada,
to Korea, Japan, there is appropriate regulation in place.
Depending on the jurisdiction,
it's a mix of old rules, new rules, binding, non-binding,
finance-specific, cross-sectorial,
given the technology neutral approach,
but also that there may be a need to continue to monitor
given the rapid developments in this area
to identify and address any gaps.
So what's your view and expectation on that front?
I think it has been said, this is a fast evolving matter
and we are still lacking proper data on risks
and risk mitigation measures that are being put in place.
So I think we should not assume
that financial market participates
do not properly mitigate risks associated with AI.
Now, we have specific risks in AI
that are not covered right now
by the current rules,
explainability, AI-driven market distortions.
We have this risk of the black box
of the lack of accountability, transparency.
We have the risk of reinforcing biases
and we have the risk of flawed algorithmic trading
and model drift.
So we do have specific risks of the use of AI in finance
that are not currently being addressed.
Probably the largest one is the one
that Gary Gensler flagged on the herding risk
and the reliance on a few data models
that Tom also mentioned earlier on
that could potentially lead to a financial stability risk.
My view is that, first, we are not fully equipped
at this point in time to make a call
whether or not we need specific regulation
for AI in finance in addition
and over and above to the AI Act.
And second, from a competitiveness point of view,
if we came to the conclusion that we need something else,
it should be at international level and not at EU level.
Because let's face it,
all systems widely used today are US systems.
They are the generalist systems,
the generative AI systems that I mentioned,
they are the Bloomberg GPT, they are the Bond GPTs,
they are all American systems
and tomorrow there will be American, Indian, Chinese systems.
So if we wanted to have bespoke AI
for finance regulation in Europe,
the risk would be to build fortress Europe
and to prevent financial market participants
from properly using the systems
that are being deployed outside of the European Union.
So what we then would need is international regulation
sponsored by the FSB, sponsored by IOSCO,
and one good example of how we could go about this
was the Bletchley Declaration in November 23 in the UK,
where the European Union plus 28 other countries
came together and looked at the risks of AI,
also AI in finance,
and were thinking about what framework do we need,
and this comprised China, this had the United States,
this had the UK,
and I think that would be the way to go.
On the supervisory front, Tom,
do you think that there is perhaps a need
for more supervisory guidance to help companies
in their compliance going forward?
We're certainly open to it, yeah.
But I think we'd need to,
and this is the point of the consortium,
we'd need to understand exactly where it's needed.
Just issuing a load of guidance
and imagining it's going to make everything clear
is not necessarily the way forward.
We've heard some specific stuff
around senior managers' responsibilities,
around model risk management policies.
We'll offer some clarifications there.
And I think for me,
one of the big things to think about actually here
is how we can use artificial intelligence
to lighten the compliance burden and improve
the risk management and compliance operations,
both of firms, but also of the institutions.
I mean, Petra raised the interesting question
about using artificial intelligence
to respond to consultation papers.
We've actually done that in the UK,
the digital pound consultation of 50,000 responses.
No way we can read 50,000 responses.
We used some computational techniques to review that.
We've published all of our approach to that.
It was legally novel,
so we had to go and get legal opinion on this.
And actually, it was exactly what Petra expected,
which was we could see all of the templates,
the letter writing campaigns,
and you could see it very, very objectively.
You could see it visually.
And actually, it allowed us to process
without human bias, with proper attention,
all of those responses,
because if not, 50,000 responses
would have been unmanageable.
So I think the more we can use these sorts of techniques
and look to the positive,
and the positive in that case
was the fact that we managed to review
all of those responses.
We did it consistently.
We did it without human bias.
And then we had mechanisms for diving into the ones
which were more distinctive.
So yeah, we're up for guidance,
but I think we should also be thinking
about how we use artificial intelligence ourselves.
So the example you're using
touches also upon skills and capacity.
And I'd like to ask both Claude and Jonathan
a very quick reaction on that front,
on the skills and capacity,
both for the supervisors and the institutions,
as well as the industry.
So I mean, I mentioned it earlier.
I think the supervisors are uniquely placed
to look at the risk across different layers
of the ecosystem and have that open.
I mean, it's a real challenge for supervisors to do that.
I think everyone appreciates that,
particularly as this is evolving so quickly,
but they are uniquely placed to be able to do that.
So it's a key role as this area evolves,
but they can only do that if everyone is talking to each other
if you're maintaining that level of dialogue.
If it becomes a very one-sided review discussion,
that's not going to happen.
Yeah, I think that supervisors are, of course,
constrained in terms of technical skills,
resources, and budget.
And I think all regulators in the world
face budget constraints,
and so we have to be nimble here.
On the positive side, though,
I think that I've seen, at least in my jurisdiction,
we've equipped ourselves with AI expertise,
with skills, with data scientists,
and also we've equipped ourselves with systems
that function in an air-gapped, secure environment.
And so a number of my fellow colleagues
in other jurisdictions are doing the same.
I think at the same time,
we have corporation initiatives that are very good.
So we exchange at European level
in the framework of the EBA,
in the framework of ESMA, IOPA.
We do exchange knowledge on who is doing what.
We can pick ideas from others.
There are also sub-tech tools that are developed together,
for instance, at the level of the European Central Bank,
a number of sub-tech tools have been developed
that are deployed amongst all the member states.
We have the BIS initiatives that are useful for all of us.
And so I think cross-border collaboration
and exchange of knowledge are going to be key also
to overcome the challenges that we face
in terms of resources.
So we've had a very rich discussion.
So to help our audience have some key takeaways,
I want one 30-second key takeaway from each of you,
starting with Tom, please.
These technologies are global.
They don't really recognize borders
in the way that conventional financial services
regulation does.
So how international cooperation is going to work,
thinking about things like Bletchley,
Paris Summit coming up,
but then also the work of international organizations,
OECD, Financial Stability Board, et cetera.
That's the really key thing for me,
much as Claude mentioned.
Digital literacy and financial literacy
must be our key priorities.
If we're having an honest and true conversation
about competitiveness, we should all know
at the end of the day, it's all about scale.
I mean, for me, I think it's just collaborating
to make sure we get the right regulation
to manage the risk whilst continuing
to support innovation and competition as well.
On the market side, I would say collaboration is key.
I would also add a word on the corporate side.
We are missing talent at board level right now,
and we are missing talent at the C-suite,
senior executive level,
and I think those are going to be key going forward.
Thank you, everyone.
It was a fascinating discussion,
and please welcome me.
Please, let's thank all our speakers.
Thank you.
