A Spanish Regulator Logged a Data Breach Run by an AI Agent. What It Means for a Small Business.

On 14 September, Spain's data protection regulator, the AEPD, published details of a breach notification involving an AI agent. According to its deputy director's account, the agent scanned files for weaknesses, logged in, moved through the system looking for more flaws, changed personal data and reached invoices (
).

What is, and is not, known

The AEPD calls it the first notification of this kind that it has received. It has not named the organisation, the attacker, the AI model or the number of people affected, and the case is still under review. Reports differ on who set the agent going: one says a third party pointed a well-known model at a Spanish organisation, another says the affected organisation described the agent. So "an AI acted entirely alone" is more than the regulator has said. Its main point is speed: AI does not create new kinds of attack, but it can make them faster and easier to repeat.

The same fortnight, in Australia

• Treasury's advice to the Treasurer, reported in early September, is that about two-thirds of Australian businesses use AI in some form, but fewer than one in ten call that use significant (
  ).
• The government opened consultation on proposed mandatory national AI standards, aimed at data centres and the training of frontier AI models. It closed on 9 October, and the government plans to legislate early in the new year (
  ).
• The Finance Minister asked the Public Service Commission for advice on whether smart glasses should be restricted in Commonwealth workplaces. Nothing has been decided (same ABC report).
• On 17 September, King Charles told a gathering of AI leaders in Scotland: "Surely we need sufficient means of control before it is all too late?" Attendees included the heads of Nvidia and Google DeepMind and OpenAI's chief financial officer (
  ).

The standards under consultation are aimed at the big end of the industry, not at a small business using a chatbot. But the direction is clear.

The question for you

You are not running a frontier lab. The lesson still holds at your size: if any AI tool has access to your customer data, your inbox or your invoicing, do you know what it is allowed to do without you watching? Most owners who have connected one have not asked, because it was switched on quickly by whoever set it up. It is a twenty-minute question if you look.

Find out what is connected in your business →

Updated 16 October 2026 after re-checking every source.