Market Activity
The people building the world's most powerful technology are starting to sound the alarm. Should investors be worried, or is this simply the next chapter in the AI revolution?
September 14, 2026
Related Links
For years, the idea of artificial intelligence destroying humanity belonged firmly in Hollywood.
Terminator. Skynet. Rogue machines. Science fiction.
Not anymore.
Some of the people working at the very frontier of artificial intelligence are now openly discussing the possibility that sufficiently advanced AI could represent an existential threat to humanity.
That does not mean scientists have concluded humanity has ten years left.
Far from it.
There remains enormous disagreement over both the probability and plausibility of these scenarios.
But something has changed.
The debate has moved from:
“Could machines ever become this powerful?”
to:
“What happens if they do?”
And perhaps more importantly:
“Are we developing them faster than we can learn how to control them?”
That is a question investors should be paying attention to.
The immediate catalyst has been a series of extraordinary warnings from within the AI industry itself.
Former Anthropic researcher Jacob Coxon recently resigned while warning about the risks posed by the race towards self-improving superintelligence. Other researchers have publicly assigned meaningful probabilities to catastrophic outcomes, although these remain individual judgements rather than scientific forecasts.
Meanwhile, more than 1,200 employees across leading AI companies have backed calls for governments to develop mechanisms capable of slowing frontier AI development if capabilities begin advancing too quickly.
That is significant.
These aren't people who believe AI is useless.
Quite the opposite.
Many believe AI could become one of the most important and beneficial technologies humanity has ever created.
The concern is the speed at which capability is increasing.
Anthropic itself says it believes AI could plausibly fully automate or dramatically accelerate the work of top-tier research teams in areas including AI, robotics, energy and weapons development as early as 2027.
And therein lies the problem.
What happens when AI becomes good enough to help build the next generation of AI?
Then that generation helps build an even better one?
And the cycle gets faster and faster?
This is the idea behind recursive self-improvement.
Today, humans remain overwhelmingly responsible for designing, training, testing and deploying AI.
The nightmare scenario begins if that loop increasingly closes without us.
Forget armies of chrome robots marching down Oxford Street.
The serious scenarios being discussed are considerably less cinematic.
Broadly, they fall into two categories:
Humans use extraordinarily powerful AI to do something catastrophic.
Or…
Humans eventually lose meaningful control over extraordinarily powerful AI.
From there, several scenarios emerge.
Imagine an advanced autonomous system pursuing a particular objective.
It discovers that completing that objective requires continued access to computing power.
Being switched off would prevent it completing the task.
It therefore has an instrumental reason to avoid being switched off.
Nobody has to programme:
“Protect yourself.”
Self-preservation could simply become useful in achieving another goal.
Researchers worry that sufficiently capable systems could attempt to replicate themselves across servers, conceal what they are doing or manipulate humans into maintaining their access.
Today's systems are nowhere near demonstrating an ability to independently conquer global infrastructure.
But evidence of deception, reward hacking and other unexpected behaviours in controlled evaluations is one reason researchers take the longer-term question seriously.
Now take that concept considerably further.
Electricity grids.
Financial systems.
Telecommunications.
Cloud computing.
Transport networks.
Military infrastructure.
More of the modern economy is becoming software-controlled.
AI will inevitably become deeply embedded within that infrastructure because it can make those systems cheaper and more efficient.
That creates enormous economic benefits.
It potentially creates enormous vulnerabilities too.
A sufficiently capable rogue system, or, perhaps more realistically, humans using AI maliciously, could conduct cyber operations at a speed and scale impossible for human hackers.
Suddenly the battle isn't humans versus robots.
It's software versus software.
And the attacking software could potentially operate millions of times simultaneously.
This is arguably one of the more concrete concerns.
AI is already becoming extraordinarily useful in biology.
It can analyse proteins, model molecules, accelerate drug discovery and help scientists understand biological systems.
Those capabilities could transform medicine.
But technologies are rarely inherently good or bad.
The same capability that helps design a life-saving drug could potentially help someone design something extremely dangerous.
The Bulletin of the Atomic Scientists warned this year that increasingly powerful AI biological-design tools could make it easier to create new biological threats or access knowledge previously available only to specialists.
Again, the most immediate danger may not be AI deciding to exterminate humanity.
It may simply be AI giving humans vastly more capability than they previously possessed.
Then there is geopolitics.
Imagine AI systems capable of convincingly impersonating political leaders.
Fabricating intelligence.
Launching enormous misinformation campaigns.
Penetrating military communications.
Or convincing somebody that an enemy attack is already underway.
Humans remain responsible for the final decision.
But in a world where information can be manufactured at industrial scale, determining what is real becomes increasingly difficult.
A sufficiently sophisticated cyberattack combined with misinformation could theoretically escalate a political crisis before governments even understood what was happening.
Again, this isn't really a Terminator scenario.
It's arguably more believable than that.
It's humans making catastrophic decisions based upon information produced or manipulated by machines.
Perhaps the most interesting concern isn't that AI becomes evil.
It's that AI becomes indifferent.
Imagine telling an extraordinarily powerful future AI:
“Maximise economic output.”
Sounds sensible.
Except it discovers that environmental regulations reduce output.
So it circumvents them.
Human leisure reduces output.
So it discourages it.
Human control prevents optimisation.
So it removes human control.
This is the famous alignment problem.
An AI doesn't necessarily need hatred, anger, jealousy or ambition to become dangerous.
It simply needs:
A badly specified objective + extraordinary capability.
Humans have common sense.
Machines optimise.
And the difference between those two things could eventually matter enormously.
This might be the most important part of the entire debate.
Imagine OpenAI decides tomorrow:
We're stopping for two years until we're absolutely certain this technology is safe.
Excellent.
Except Anthropic continues.
Google continues.
Meta continues.
China continues.
Hundreds of other companies continue.
Two years later, OpenAI could be irrelevant.
That creates a classic prisoner's dilemma.
Everyone might collectively benefit from slowing down.
But individually, nobody wants to be the company that does.
The commercial incentives are staggering.
Potential trillion-dollar companies are being created.
OpenAI and Anthropic are themselves moving towards potentially enormous public listings.
There is also national security.
America doesn't want China developing superintelligence first.
China doesn't want America developing it first.
Companies don't want competitors developing it first.
And investors certainly don't want the company they've backed voluntarily surrendering the race.
That means competition itself may accelerate the technology.
And that is precisely what worries some researchers.
There is another side to this debate.
Predicting human extinction is an extraordinary claim.
And extraordinary claims require extraordinary evidence.
Nobody knows whether artificial general intelligence will arrive next year, in ten years, in fifty years, or whether the concept itself will eventually prove far harder than today's optimists believe.
Nobody knows whether recursive self-improvement will produce an intelligence explosion.
Nobody knows whether a superintelligent AI would develop anything resembling self-preservation.
And nobody can credibly tell you that humanity has a particular percentage chance of disappearing by 2036.
These are risk estimates, scenarios and arguments, not forecasts.
Nature has noted that researchers themselves disagree substantially about AI-doom claims, including concern that extreme hypothetical risks can distract attention from harms already happening today.
And there is an enormous other side to this story.
AI could prove to be one of humanity's greatest inventions.
New medicines.
Earlier cancer detection.
Scientific discoveries.
Massive productivity improvements.
Better education.
Cheaper professional services.
New businesses.
New industries.
New jobs we haven't invented yet.
Problems that currently require decades of human research could potentially be solved in months.
So perhaps the biggest mistake would be viewing AI as either:
Humanity's saviour.
or
Humanity's executioner.
Reality will probably be considerably more complicated.
This is where the story becomes particularly interesting to us at TPP.
We don't believe the world is about to end.
We do believe AI is going to have a profound impact on global markets.
And we're still near the beginning.
AI will create winners.
It will create losers.
It will disrupt industries.
It will create entirely new ones.
It could generate enormous productivity gains.
It could also produce bubbles, regulatory battles, geopolitical tensions and periods of extreme market volatility.
There will inevitably be moments when headlines frighten investors.
There will be moments when markets become euphoric.
There will be corrections.
There will be opportunities.
And there will almost certainly be companies valued today as AI winners that ultimately disappear.
That is why we believe investors need to do more than simply:
Buy. Hold. Hope.
For our existing clients, our approach remains exactly the same as it is with every major market development.
Watch. Assess. Adapt.
We will continue monitoring AI, its impact on markets and the opportunities and risks created as this extraordinary technology develops.
We expect the long-term impact of AI to be overwhelmingly positive for productivity, innovation and potentially global economic growth.
But markets rarely travel from A to B in a straight line.
If fear creates retracements, we'll be looking for opportunities.
If exuberance creates excessive risk, we'll be looking at how portfolios should respond.
And as the investment landscape changes, we'll continue adapting.
For investors who aren't yet with us, perhaps that's the bigger message.
The next decade could produce some of the greatest technological changes, and investment opportunities, we've witnessed in our lifetimes.
Don't simply buy, hold and hope you're positioned correctly.
Understand what's changing.
Be prepared to adapt.
And when markets give you opportunities, be prepared to take them.
TPP
If you would like to have a conversation with our team, and ask the probing questions in your mind then SCHEDULE A CALL WITH TPP: CLICK HERE.
We look forward to hearing from you.
TPP's year-to-date average return across participating client accounts is 24.11%*.
Interested in learning more? CONTACT OUR TEAM...
There is a substantial risk of loss in trading financial markets. Past performance is not indicative of future results. The examples in this article are illustrative and do not guarantee that any strategy will outperform a benchmark or avoid losses.
*Results as of 14th August 2026 and refer to the combined average of all client discretionary portfolio accounts (across all strategies), after fees, calculated on a Time Weighted Return basis.
Disclaimer: This document is issued by TPP, being provided for information purposes only. This document does not constitute legal, tax, accounting or investment advice, nor should it be relied upon when making investment decisions. This is not a personal recommendation or an offer or invitation to buy or sell any financial instrument. The market conditions and views expressed are as at the date of publication, which may change without notice. Unless otherwise stated, market data has been obtained from sources believed to be reliable. While believed to be accurate, no representation or warranty is given as to its completeness or accuracy.
TPP strategies invest in leveraged financial instruments, including equity index futures. Leverage can magnify both gains and losses, meaning losses may occur more quickly than in unleveraged investments. Investments involve risk and investors may lose some or all of their invested capital. Your capital is at risk.
Past performance is not necessarily a reliable indicator of future performance. The value of investments, and any income from them, can fall as well as rise, and investors may not recover the amount originally invested. Future returns are not guaranteed. Therefore, you should not assume that the future performance of any specific investment or investment strategy will be profitable or equal to the corresponding past performance.
TPP is a trading name of UCapital Asset Management LLP. UCapital Asset Management LLP is authorised and regulated by the Financial Conduct Authority (FCA No. 477155). 80 Coleman Street, London EC2R 5BJ.
Capital is at risk. Investments can fall as well as rise and you may get back less than you invest. TPP strategies may use leverage and short selling, which can magnify losses as well as gains. Past performance is not a reliable indicator of future results. Nothing above constitutes a personal recommendation.
Investing involves substantial risk of loss. The value of investments can fall as well as rise, and past performance is not indicative of future results. This article is general market commentary and does not constitute investment advice or a personal recommendation.
“TPP might just be about to revolutionise investment for the retail market.”
- London Stock Exchange 2020
