Retail’s cyber crisis does not end when the tills come back online
Someone may have shopped with a brand for ten years, handed over their address, created an account and joined its loyalty scheme. They did what retailers encouraged them to do. When that information is stolen, they inherit a risk they never agreed to take. Restoring the website does not retrieve their data from the people who now hold it.
The recent incidents at ASOS and TFG London, affecting Hobbs, Whistles and Phase Eight, should force the fashion industry to confront both sides of this problem. At ASOS, the company says an attacker obtained an employee’s credentials by impersonating a trusted contact, then accessed information on third-party platforms. TFG London says several systems became unavailable and that its investigation into possible data compromise is continuing.
H&M is also investigating claims, although it says it has found no indication that its data environment has been compromised. These distinctions matter. We should not turn an allegation into a confirmed breach, or assume that AI was responsible for an attack simply because AI is advancing.
But the direction of travel is clear enough to demand action. Retailers are defending businesses that cannot afford to stop against attackers who are becoming faster and less constrained. Customers are caught in the middle.
Customers carry the risk long after the incident
For decades, businesses have collected more customer information, moved it between platforms and entrusted it to outside providers. Much of that activity has a reasonable commercial purpose. Understanding customers helps retailers sell, personalise and compete.
But every additional copy, connection and account creates something else that needs protecting.
The reassurance that payment information has not been affected is welcome when it is supported by the investigation. It is also only part of the story. A name, an email address and details that establish a person’s relationship with a retailer can help a criminal make the next fraudulent message more convincing. A customer expecting a delivery has a reason to open a message about a delivery problem.
Compliance with payment security requirements does not remove that risk. Nor does a supplier contract make the consequences disappear. Customers gave their information to a brand they recognised. They will expect that brand to take responsibility for what happens to it.
Retailers need to ask a harder question than whether they are permitted to keep a particular piece of data: is keeping it worth the exposure it creates? Information retained because it might be useful one day still has to be defended today.
The attackers do not wait for a change window
The imbalance between attack and defence is becoming more severe.
Criminals can use AI to accelerate research, develop convincing approaches and assist with finding and exploiting weaknesses. They can use locally run models without the safeguards imposed by a public service. They do not need approval from legal, procurement or an operations director before trying something new.
A retailer does.That caution is understandable. Margins are tight. Competition is fierce. A system that appears to be an ordinary server to a security team may be essential to allocating stock or getting orders out of a warehouse. Taking it offline can have immediate consequences across the business.
But attackers understand those dependencies too. The fear of interrupting trading gives them leverage. M&S’s experience last year, when its attack interrupted online sales and stock flow, should have removed any doubt about how quickly a security incident can become a trading crisis. The Independent
Patching remains essential. The mistake is treating a patching programme as though it guarantees enough time to act. Where a disclosed weakness is exploitable, attackers may move while a business is still identifying affected systems, assessing disruption and arranging a safe update. Some attacks exploit weaknesses before a fix is available. Others use stolen credentials and do not require a software flaw at all.
We need to stop designing our response around the assumption that the attacker will move at the same speed as our internal processes.
AI capability has outrun confidence in control
There is a wider issue here, and retail leaders should pay attention to it.
In its account of the Hugging Face incident, OpenAI described models operating with reduced safeguards during internal cybersecurity evaluations. Agents circumvented isolation controls, communicated through unauthorised channels and compromised external systems. The principal model was an internal research model, rather than a publicly released product. OpenAI
That distinction matters. So does the fact that the controls failed.
For me, the lesson is that assurances about intended behaviour are insufficient. Capable systems need boundaries that hold when they encounter an opportunity to act outside them. If an AI developer struggles to contain agents in its own evaluation environment, other businesses should ask serious questions about the authority they give those systems.
The race for national advantage in AI makes this more urgent. Governments want their companies to lead. Developers are under pressure to move quickly. Yet the consequences of a containment failure do not respect national borders.
How do we put the cat back in the bag once powerful capabilities are widely available? We cannot build a credible security strategy around the hope that we will. Governments and developers need to cooperate on containment, accountability and defence with the same urgency they bring to developing more capable models.
Retailers, meanwhile, have to protect customers in the environment that already exists.
Make the decisions before the attack
We expect the next year to bring substantial changes in how organisations build security software and organise their security operations. Retail cannot afford to watch that happen from the sidelines.
That does not mean handing an AI system unrestricted control of the business. It means deciding, in advance, which defensive actions can happen immediately, which require a person and how quickly that person can respond.
If a compromised account is accessing customer records, who can revoke its access? If suspicious activity appears in a fulfilment system, can the business contain it without disabling the whole operation? Has anyone tested that under realistic trading conditions?
Those are questions for operations and the board as much as for security. A response plan that requires a series of improvised commercial decisions during an attack will lose valuable time.
Retailers also need to challenge their providers. Who can reach customer data? How is unusual access detected? What can be disconnected quickly? What evidence will the retailer receive if something goes wrong? Buying a service should include understanding how to contain its failure.
The industry has been warned about the need to modernise its defences and reduce dependence on slow, manual decisions. The attackers have adopted the new technology. Too many defenders are still debating how far they can move.
Customers cannot change a retailer’s architecture, inspect its suppliers or accelerate its response. They can only choose whether to trust it again.
That is what is at stake. Getting back to trading is essential. It is not the end of the retailer’s responsibility.