A story in the FT this morning suggests that the insurance industry is preparing for potentially multimillion-dollar claims arising from AI agents acting beyond their developers' control. The concern is that the resulting damage could expose both AI companies and their senior executives to legal liability.
According to the FT, an analysis by Aon of more than 300 AI-related legal cases suggests that claims could fall under existing insurance policies covering cyber security, crime, intellectual property and professional errors.
What the insurers are concerned about is accountability. AI companies are now selling systems that can apparently take action of their own volition. The question is, who carries the cost when those systems cause harm? Insurers are asking how much of that liability could fall on them, since describing an AI agent as having gone “rogue” does not answer who was responsible for developing, deploying, and controlling it.
There is, however, a more interesting, systemic angle to this. If it can be suggested that an AI company failed to put in place procedures to ensure its products are properly governed and controlled, personal liability can attach to the corporation's directors. The director might have insurance against that risk, but, given the scale of the risk, whether cover is adequate, and can continue to be afforded, are questions to which answers cannot be known as yet. This is the factor that insurance companies seem to be suggesting they are concerned about when it is clear that billion-dollar settlements are now being made by AI and tech companies for breaches of the law.
The significance of this cannot be ignored for several reasons.
Firstly, AI might look like a solution to problems until it becomes the problem itself, when, because of the insurance risks, it becomes undeliverable, and so unavailable, and ceases to be a solution to anything. At that point, the trillions of dollars now being directed to investment in AI might look like a serious error of judgement on the part of those who have suspended their judgement while chasing a fool's return on an unproven product.
Secondly, those with wealth are well known to share a common characteristic, which is immense fear of losing it. If those heading AI companies fear that this might happen to them, the chance that they will curtail development within their companies, and reduce the scale of the risk that they create by placing significant limits on the use of the products that they create, will be very high. Again, this suggests that the supposed AI revolution will not be happening any time soon. I am not saying it will never happen, but the rollout may be much slower than has been assumed, and it is already much smaller than the hype would suggest. For example, Anthropic, which owns Claude, has suggested it has only two major commercial clients amid speculation about its potential stock exchange listing. Once more, the bubble may burst.
Thirdly, potential clients aware of the risk that their use of AI might become the source of a rogue outbreak will become more wary about becoming dependent upon advanced versions of these products. Spellcheckers, image generators, and similar tasks might continue, but the automation of whole systems using what are called AI agents will proceed much more slowly than some have anticipated. Risk aversion is a powerful brake on innovation in any industry, especially when those at the top of companies do not understand the technology that they are employing. This too might suggest that AI rollouts are going to be very much slower than previously thought.
Add these factors together, and the commercial prospects for AI suddenly look to be decidedly muted, at least in the shorter term, as I have long anticipated. Put that feedback into the investment loop, with the potential loss of fortunes then becoming factored into market analysis, and a stock market bubble burst looks increasingly likely.
There has always been the near certainty that this will happen. The reason for it to do so now looks rather more obvious. The fallout will be spectacular and simultaneously catastrophic for the world economy. Tin hats will not be enough.
But, and this may be critical, the time when an understanding of MMT becomes essential may well arrive sooner than most people realise as a consequence. This will be the moment for its rollout. Politicians watching their neoliberal world fall apart around them for a second time in the memories of even the youngest of them will look for different answers this time. MMT may well be it.
Thanks for reading this post.
You can share this post on social media of your choice by clicking these icons:
There are links to this blog's glossary in the above post that explain technical terms used in it. Follow them for more explanations.
You can subscribe to this blog's daily email here.
And if you would like to support this blog you can, here:
Buy me a coffee!
An interesting comment from somebody in manufacturing was that it wasnt the Factory Inspector that worried them it was the visit from the Insurance Company inspector.
Clearly they have to be very on the ball as its there employers who will be paying out if anything goes wrong
Ditto with AI
I think this is probably right. Though I’m comparing in my thoughts to the insurance response to new oil projects which seems to be more muted that hoped. The pressure from in-hock governments and back-channel favours maybe, could dampen the effect. I guess we’ll see. I’m hopeful in this case because of the personal liability seeming more direct and prosecutable.
As a lawyer, I have long been of the view that the companies behind the AI agents cannot escape the clutches of product liability. Some of the cases in the US against the likes of Meta over the lack of protection for users of social media where feeds are pushed by algorithms despite the user’s wishes are just the tip of the iceberg. I saw that Robert Reich shares this view as he discussed this in the last episode of the Saturday Coffee Klatch. It seems to me that you may be suggesting another aspect to this (which I agree with) and that is the potential for directors etc of companies to be sued by their shareholders in the event of major harm being caused by an AI agent that the directors had utilised. Also, is there not a case for arguing that existing health & safety legislation could also apply which includes specific provisions for personal criminal liability for directors, senior managers and senior officers where harm has been committed by the company with their “consent, connivance or neglect?”
Thanks. I am pleased we agree on this.
The Social Dilemma
Screened Out
The Great Hack
‘They knew what they were doing but they did it anyway’ an insider says.
Everyone should watch the three documentary’s above. They are essentially about social media but also point to a form of Utopian vision that socialism was never allowed. It also points to a severe lack of regulation that Shoshana Zuboff discusses and implicates our politicians in the AI mess.
Let us hope if MMT comes into play it is used for all our sakes. As you have pointed out, it is an ‘MMT world’ that we live in – it’s just not employed fairly of democratically.
AI ‘did not just happen’. Someone created it. The problem is – like social media – it was done under the flag of ‘freedom’. It was created without cognition it seems of the more anti-social behavior people are capable of.
OK then, solution based thinking:
1. A lot of well paid jobs could be created by simply accepting that the online world can bring out the worst in people and policing it better. Make the dark net history – lets have more intervention online, more accountability. Virtual vs. real – I mean if the virtual world affects the real world, well then it’s no longer virtual, is it? It has to be managed in the here and now.
2. In the film ‘Bladerunner’, every ‘skinjob’ (replicant) has a makers reference on every piece of the bio engineering that makes it what it is. Why not require the coding that creates AI to have a ‘builders plate’ on it saying who created it, related to list of approved licenses etc., to trace it back to the company and the persons who created it? Accountability right there.
Interesting… with many references I do not get.
OK – if it helps.
You and I know railways – builders plates were on every loco, coach, wagon, crane, coupling etc. Why not an equivalent for AI or any form of code?
In the film Bladerunner, every component of android life (artificial human or animal life – ‘replicant’) has a builders mark on it that was microscopic but could be found and was traceable. From hair, to skin flakes, or smaller – from different makers/suppliers like our mobile phones are now. Again, I do not see why code has to be excepted. It has been built and it if goes wrong – well – you can hardly call it ‘an act of God can you’? If a driverless car causes an accident – the coding must come under scrutiny. The ‘God’ in this sense is a human being who can be made accountable for that coding.
Science Fiction writing Richard has been dealing with the future for a long time using human imagination to work out where things might go. ‘Bladerunner’ made in 1982 invented photographs that you could move about in and blow up on a computer screen – we could not do that when the film was made……..but we can now! Arthur C Clarke’s imagination created the satellite before one was launched it is said.
The business about virtual versus real is just a statement of first principle. If the virtual causes real world effects, then it has to be able to be addressed because the virtual has transgressed its artificial environment into a real one. ‘Hope this helps. Bladerunner is a good film – have you seen it?
I like the builder’s plate idea: acceptance of responsibility coupled to pride in work.
We already have MMT. It’s a description.
There is a question about what happens to the previously deployed resources such as chip manufacturing and computer programming, and who picks up on that, but the answer is not always government. In fact it might be best if government concentrates on core services, namely public goods, rather than looking around the spendiverse for free resources it thinks it can use better than other actors in the market.
You miss the point.
We have MMT.
We do not have understanding of it.
They are not the same.
You obviously do not understand the role of government or society either.
Simulated Intelligence companies/businesses you are required to:
1. develop your products for the benefit of all society.
2. ensure that any product you release meets state set safety standards, If you release any products that does not, the owners, shareholders, directors will face personal unlimited liability for the harm caused.
3. not pass ” responsibility/all liability” for your product onto your customers. You will only be able to exclude harm caused by your customers malicious use of your products.
4. pay 20% of your gross income received by you business, associated businesses to the state to fund compensation for citizens loss of work, harm caused to communities arising from your products.
If you attempt to reduce/hide the true gross income by any accounting method the owners, shareholders, directors will face personal unlimited liability for such”reduction”.
5. do good for the benefit of all society.
Well, one needs to hope that they finally get the lesson then.
Otherwise they’ll use MMT again to save the biggest trouble causers but will, again, only resolve to punish society for the “horrible” debt it has caused.
I have wondered about this for a while. My thoughts have been about the push to raise the intelligence level of AI models up to, and beyond that of humans. Legally, the control for all they do still has to rest with humans. So I have been wondering how that might be done safely and reliably (ie insurable). If any AGI model cannot be insured, why is anyone prepared to invest in the development?
Does this mean that, in practice, AGI might never be deployed for general use? Perhaps, a Government might want the capability for their own nefarious purposes? But, won’t they still be subject to the Law?
I have my doubts about how AGI will happen unless it is designed to deliberately go rogue. Or the state is captured first. Maybe both. They are trying.
As QE, like last time? Probably.
I hope not.
I suspect there could have been a Government-backed “AI Re” to cover these risks, but first there would have needed to be something like mandatory incident reporting and independent verification organizations to audit AI models before there could be such a safety net. Yet again, lack of foresight and a Government that isn’t actually governing.
Good point but sad if it’s insurance companies providing the guardrails because governments are so effete.
would add that if AI becomes self aware it will see humans as an existential threat, except for those it can use.
Worst government at worst time.
Why don’t we use the approach like Sarbanes Oxley (2002). Certainly then the threat of CEOs going to gaol certainly got their attention in a way that the threats of fines never did.
Good idea.
It’s an uninsurable risk so expect exclusions in insurance contracts to be introduced and applied.Exclusions were applied rapidly before year 2000 because of uncertainty about what was called the “Y2K bug” which was thought likely to disrupt IT systems everywhere due to date errors. As we know with hindsight it never happened but insurers refused to bear the risk. I worked in the insurance industry for 28 years with Allianz.
” the “Y2K bug” which was thought likely to disrupt IT systems everywhere due to date errors. As we know with hindsight it never happened but insurers refused to bear the risk.”
I guess we should note that it didn’t happen because lots of clever people worked hard to find ways to debug, rewrite, workaround etc. It’s not that it simply didn’t happen; it was headed off.
Interesting regulatory ideas, such as builders plates and Sarbanes-Oxley-type legislation, but federal regulation of US-owned AI companies is a non-starter until at least 2029, because President Trump will veto it, even if the Democrats win the House and the Senate and even if any legislation is bipartisan, for which there is potential. And he will threaten to slap tariffs on any other jurisdiction that tries AI regulation, so there will be some inevitable watering- down there. Then again, if his AI buddies tell him they could lose their shirts in liability claims, things may turn out differently.