State of play, 22 September 2026. The Plan IA360 was presented on 21 September. There's no call for applications and no governing rules a business could apply to today, for any project in the plan. What follows doesn't depend on any of them ever being rolled out: it's about a risk that already exists today, with or without the plan.
Among the projects the plan announces is an AI literacy strategy for the adult population that includes, among its content, «identificación de contenidos sintéticos y de fraudes» ("identifying synthetic content and fraud"). That phrase describes the problem well, but at the scale of a country. At the scale of a specific business, synthetic content is no longer a digital-literacy matter: it's a very concrete route for fraud, aimed at its brand or at the people who work there.
A cloned voice that sounds like the CFO's, asking by phone for an urgent transfer. An email that imitates management's writing style to the point that nobody questions it before clicking. A generated image showing a product with a defect that never existed, circulating before the business even knows it's out there. These are three fraud patterns that already circulate, with the difference that generative AI has lowered the cost and raised the credibility of pulling them off well.
Why this isn't only a classic cybersecurity problem
Traditional email filters and verification protocols were designed to detect technical signals: a fake domain, a suspicious link, wording with errors that gave the fraud away. Synthetic content doesn't need any of that. A call with a cloned voice carries no technical indicator a filter can read, because it arrives as voice, in a conversation that sounds like the real person. The problem has shifted from what a system detects automatically to what a person has to know how to recognise.
The real defence against a cloned voice or email is procedural
Real-time synthetic-voice detectors exist, but they're designed for large-scale call centres, not for an executive's mobile phone, and it isn't wise to trust them alone with the decision to authorise a payment. The defence within reach of any business, without depending on that technology, is procedural, and it costs little to put in place:
- Double verification of any unusual request — a change of bank account, an urgent payment, skipping a usual step — before carrying it out.
- An internal codeword, agreed in advance and known only to whoever needs to use it, to confirm sensitive instructions received by voice.
- A second, pre-agreed channel to confirm any payment or change of banking details: never the one proposed by the call or the email itself.
These three measures don't depend on any provider or any technology budget. They're the ones that most reduce the risk of a voice-impersonation fraud, and it's worth having them active before looking at what technology can contribute on two different, more narrowly defined fronts.
Two fronts where technology does help, each in its own lane
Neither of the following two services stops someone cloning an executive's voice with public audio: the process above is what stops that. What they do is close two other doors, each in a different area.
Making sure the business's own chatbot or agent doesn't say what it shouldn't. An assistant connected to the brand — the customer-service chat, an agent replying on social media — can be manipulated with jailbreak techniques into saying things the business would never approve, with its name in front of the customer. Checking that before a third party does is the job of Summum IA's AI Red Teaming: controlled, reproducible attacks on the business's own models and agents, with a prioritised findings report. Among the risks it targets is the one its own page names: reputation damaged by a viral jailbreak.
Making sure internal information doesn't leave without control. The other front is data leaking out towards AI tools nobody has authorised: any contract, customer database, source code or internal documentation pasted into a free AI tool leaves the business's perimeter without a trace. Controlling what can leave, and with what record, is the territory of Summum IA's corporate AI Gateway: visibility over which tools the team uses, filtering of what sensitive information can leave through them, and a record of every use.
Signs that should set off an alarm
- A request for a transfer, a change of bank account or an urgent payment by phone or audio, without confirming through the second channel already known in advance.
- An email that matches the sender's usual style but asks you to skip a verification step that's normally never skipped.
- Visual content of the business (product, premises, staff) circulating without anyone on the communications team having generated or authorised it.
- Unusual urgency as the main argument: fraud by synthetic content works best when there's no time to check anything.
Frequently asked questions
How do you tell a real call apart from one with a cloned voice?
Today, the reliable route is confirming through a second channel already known in advance any unusual or urgent instruction received by voice or email, never through one proposed by the call or the email itself.
Does AI Red Teaming stop a customer receiving a call using a clone of my director's voice?
No. Red teaming tests the business's own chatbot or agent; it doesn't protect against a third party cloning a voice with public audio. Against that, the defence that works is the double-verification procedure described above.
What do you do if someone has already impersonated the business's image or voice?
If a payment was involved, the first step is alerting the bank to try to hold it. After that, you document the available evidence (recording, number, emails), notify the platform where the content is circulating, report it to the police, and review internally what real material could have been used as the basis for generating it. INCIBE (Spain's National Cybersecurity Institute) also assists businesses on 017, its free cybersecurity helpline.
Do I need both services, or is one enough?
They cover different risks. The AI Gateway limits what material leaves the business uncontrolled from the inside; AI Red Teaming checks whether the business's own assistant can be manipulated from the outside. Which one is needed first depends on whether the business already has a public-facing chatbot or agent, or on whether the bigger risk today is internal material leaking out.
The rest of the Plan IA360 is in the complete guide to the Plan IA360.
Flagship project 6 (AI literacy strategy), Education area, Plan IA360, La Moncloa, 21 September 2026. Full official document (PDF).
Last verified: 22 September 2026.