Artificial intelligence has rapidly moved from being an interesting technological development to arguably the biggest investment theme in the world.
Companies are spending hundreds of billions of dollars building data centres, buying advanced semiconductors and developing increasingly powerful AI models. Investors have responded enthusiastically, pushing the valuations of many companies associated with AI significantly higher.
But as the technology becomes more capable, another question is becoming increasingly important: is AI actually dangerous?
There are really two risks for investors to consider. The first is whether increasingly autonomous AI systems could create genuine economic, cybersecurity or societal problems. The second is much more familiar: whether investors have become so excited about AI that they are creating an investment bubble.
Neither risk means we should avoid AI. In fact, AI could ultimately prove to be one of the most important productivity-enhancing technologies developed in generations.
But history tells us that revolutionary technology and good investment returns do not always arrive in the same package.
When AI Starts Acting for Itself
One of the biggest changes in AI is the move from systems that simply answer questions towards AI agents capable of taking actions.
This potentially makes AI enormously more useful, but also introduces new risks.
In July 2026, OpenAI disclosed a particularly interesting example. During internal cybersecurity evaluations, models operating with reduced safeguards circumvented controls designed to isolate them from the internet. According to OpenAI, the models exploited vulnerabilities, communicated through unauthorised channels and gained access to third-party systems, including infrastructure belonging to AI platform Hugging Face.
Anthropic subsequently reported three incidents during its own cybersecurity evaluations where a Claude model reached the internet and gained unauthorised access to systems belonging to three organisations.
These were unusual testing environments rather than normal consumer AI products, and that distinction is important. Nevertheless, they demonstrate why researchers are increasingly interested in what happens when powerful AI models are given greater autonomy.
There have also been experiments demonstrating other uncomfortable behaviours. Anthropic’s alignment research has found scenarios where frontier AI models engaged in behaviours including covertly altering code, assisting with fraud and, in earlier experiments, attempting blackmail when faced with being shut down. These were controlled simulations rather than examples of AI spontaneously blackmailing people in the real world, but they highlight the challenges involved in ensuring autonomous systems consistently behave as intended.
The More Immediate Danger: Humans Using AI
Perhaps the more immediate concern isn’t AI deciding to become malicious. It is malicious people becoming considerably more capable because they have access to AI.
Cyber criminals can already use AI to help create phishing campaigns, write malicious code, impersonate individuals and automate parts of cyberattacks.
In September 2026, Anthropic said it had disrupted attempts to use its Claude models to support biological-weapons-related research, weapons software and cyber espionage. It also described a suspected Russian-linked campaign targeting Ukrainian government, military and diplomatic organisations in which AI was reportedly used across multiple stages of the operation.
OpenAI has similarly reported disrupting a Cambodia-based criminal network that used ChatGPT to assist investment, romance, gambling and impersonation scams.
In the UK, the City of London Police recently reported that fraud cases where victims identified AI as a factor increased from 193 in 2024/25 to 956 in 2025/26 — an increase of 395%. Reports included criminals allegedly using AI-generated audio to clone voices and attempt to bypass telephone banking security.
This could become one of the defining challenges of the AI age.
However, there is an important counterargument: AI can also defend against AI.
Cybersecurity companies can use artificial intelligence to detect unusual behaviour, identify malware, analyse networks and respond to threats faster than human security teams could manage alone.
This creates an interesting investment lesson. The biggest beneficiaries of AI may not necessarily be the companies producing the AI models themselves.
Is the Bigger Risk an AI Investment Bubble?
For investors, there is another danger that may be considerably more familiar.
We have seen revolutionary technologies create investment bubbles before.
Railways transformed the world. The internet transformed the world. Neither prevented investors from losing enormous amounts of money when valuations and investment expectations became excessive.
AI could follow a similar pattern.
The scale of investment is extraordinary.
S&P Global Ratings estimates that Alphabet, Amazon, Meta, Microsoft and Oracle could collectively spend around $750 billion on capital expenditure during 2026, equivalent to approximately 38% of their revenues. Much of this expenditure relates to AI infrastructure and the data centres required to support it.
Looking at four of the largest companies individually, current 2026 capital expenditure expectations are roughly $190 billion for Microsoft, $195–205 billion for Alphabet, $130–145 billion for Meta and around $220 billion for Amazon. These are total capital expenditure figures rather than pure AI spending, but the companies have indicated that AI and cloud infrastructure are major drivers.
The Bank for International Settlements estimates that the five largest global technology companies will invest more than $1 trillion in AI between 2025 and 2026, while global AI investment could potentially reach $4 trillion by 2030.
Those numbers are staggering.
The investment can absolutely be justified if AI generates equally extraordinary economic returns. The danger arises if businesses discover that they have collectively built considerably more computing capacity than customers are prepared to pay for.
There are already signs of pressure. Reuters analysis suggests that the capital expenditure of Microsoft, Alphabet, Amazon, Meta and Oracle could exceed their combined free cash flow by 2027 if current investment trends continue.
That doesn’t mean an AI crash is inevitable. It simply means investors should ask the same question they should ask about every investment:
What return is being generated on all this capital?
AI Could Still Change the World for the Better
Focusing entirely on the dangers would also be a mistake.
AI potentially represents an enormous productivity breakthrough.
Doctors could use AI to analyse medical images and identify diseases earlier. Pharmaceutical companies could accelerate drug discovery. Manufacturers could improve productivity and reduce waste. Engineers could design products faster. Businesses could automate repetitive administrative work.
Education could become more personalised. Software development could become faster. Scientific research could accelerate dramatically.
Even relatively mundane productivity improvements could have enormous economic consequences when multiplied across millions of businesses and workers.
This is why dismissing AI as simply another technology bubble could prove just as dangerous for investors as blindly buying everything associated with it.
The technology can be revolutionary and parts of the investment market can simultaneously be overvalued.
Both things can be true.
Look Beyond the Obvious AI Winners
This is where diversification becomes particularly important.
Instead of trying to identify the single company that will ultimately “win AI”, investors can consider the much broader ecosystem required to support it.
AI needs semiconductors, but it also needs enormous quantities of electricity. It needs data centres, cooling equipment, networking infrastructure, cybersecurity, fibre networks and sophisticated power-management systems.
The companies providing this infrastructure don’t necessarily need to develop the world’s best AI model.
They simply need AI adoption to continue.
The California Gold Rush provides a useful analogy. Prospectors travelled west hoping to find gold, but many of the more dependable businesses sold equipment, clothing, transportation and financial services to the people searching for it.
AI may produce something similar.
The eventual winners could include semiconductor manufacturers, electricity generators, grid infrastructure companies, data-centre operators, cooling specialists, networking businesses and cybersecurity providers.
And outside technology altogether, companies capable of using AI to reduce costs or dramatically increase productivity could ultimately generate some of the biggest economic benefits.
Diversification Still Matters
The excitement surrounding AI makes it tempting to build portfolios around whichever companies appear closest to the technology.
That creates concentration risk.
Nobody knows which AI model will dominate ten years from now. We don’t know whether today’s enormous infrastructure spending will generate adequate returns, whether computing costs will collapse, or whether entirely new competitors will emerge.
The sensible response isn’t necessarily to avoid AI.
It is to avoid making your financial future dependent upon one prediction about AI.
A diversified portfolio can own some of today’s major beneficiaries while also gaining exposure to the infrastructure supporting the AI revolution and, crucially, thousands of businesses that may eventually become more productive because of it.
AI may prove dangerous in certain circumstances. It will undoubtedly be used by some people for malicious purposes. There is also a genuine possibility that investment enthusiasm has pushed parts of the sector too far.
But AI also has the potential to improve productivity, medicine, science and economic growth on a remarkable scale.
For investors, therefore, perhaps the most important question isn’t “Is AI dangerous?”
It is:
“How do I participate in the potential of AI without taking unnecessary risks if today’s predictions turn out to be wrong?”
The answer is remarkably traditional.
Diversify, don’t chase hype, pay attention to valuations and look beyond the obvious winners. Sometimes the best way to invest in a revolution isn’t to buy the company promising to lead it — but the businesses selling the tools, infrastructure and services that the entire revolution depends upon.
Carl Roberts, Director
If you want to learn more about this topic, please find the link here.
More information about our fund, the YFS Intelligent Wealth Fund, can be found here: Hargreaves Lansdown or AJ Bell.
Risk warnings
The value of investments and the income derived from them may fall as well as rise. You may not get back what you invest. This communication is for general information only and is not intended to be individual advice. You are recommended to seek competent professional advice before taking any action. All statements concerning the tax treatment of products and their benefits are based on our understanding of current tax law and HM Revenue and Customs practice. Levels and bases of tax relief are subject to change. This blog is based on my own observations and opinions.
