At M360 ASEAN in Kuala Lumpur last week, the GSMA released its ASEAN Consumer Scam Report 2026. Nearly half of reported scam cases across six markets remain unresolved. Nearly nine in ten consumers said they had recognised AI being used in scams, from deepfakes to voice cloning. Julian Gorman, the GSMA’s head of Asia Pacific, summed it up: “Access alone is no longer enough.”
A month earlier, the GSMA’s Mobile Economy Asia Pacific 2026 named the rise of AI and declining digital trust as two of four forces shaping the region’s next phase. The share of ASEAN consumers who say they have been scammed rose from 31 per cent in 2024 to 45 per cent in 2025.
Across the region, operators are now moving AI into the customer app. The technology operators are racing to deploy is the same technology eroding the trust they need to deploy it. That dependency is missing from most business cases.
Trust is now a churn number
For years, trust sat in the compliance column. That is changing. The 2026 scam report found that people who have been scammed are more than twice as likely to switch accounts or providers as those who have not. The 2025 edition found that 36 per cent of consumers would switch financial services provider for stronger security built on telecom data.
Operators hold both sides of that equation. Their networks carry the calls and messages scammers rely on, and their data can help stop them. Some are already treating protection as a product: CelcomDigi sells a dedicated safety range alongside its mobile and fibre plans, and Globe has convened a Philippine anti-scam coalition with the Global Anti-Scam Alliance. Trust is already on the price list.
The risk sits with the operator
AI inside a telco app runs on the operator’s license and on a billing relationship that may be decades old. When a vendor’s model gets something wrong, the regulator calls the operator, and so does the customer.
Southeast Asia has mature rules for networks and, increasingly, for personal data. Consumer-facing AI inside an operator’s app is newer ground, and in practice much of its governance still lives in vendor contracts. That is a thin layer on which to rest a national operator’s reputation.
Also Read: Inside Singapore’s biggest telecom cyber defence operation
Subscribers have already written part of the spec
The most useful finding in the GSMA’s 2025 scam research was not a victim count. Consumer comfort with sharing data with conversational apps fell from 53 per cent to 40 per cent. In the same survey, 72 per cent said they were comfortable with limited, purpose-bound checks that use network signals to protect them.
Read together, those numbers describe the AI subscribers will accept. Not a chatbot that asks them questions. A system that works in the background, for a defined purpose, on as little personal data as possible.
Four tests operators can apply
In this context, trust is architecture, not assurance. Four design choices can be tested before anything goes live.
- First, the AI reads patterns, not people. There is no conversational surface between the subscriber and the model, so there is nothing to impersonate and nothing to phish.
- Second, the operator holds the identity key, and the intelligence layer sees only irreversible identifiers.
- Third, data is processed in-country.
- Fourth, the capability is exposed as an operator-controlled API: every call is logged, priced and revocable, and the operator can switch it off.
We had to build to these tests ourselves, because operator security teams kept asking the right question: if this sits inside our app, how do we know you will not walk off with our data? The answer that worked was not a promise. It was something their own teams could test before launch. That should be the standard for every AI vendor in this market. Do not ask for trust. Hand over something to test.
Also Read: How Agentic AI will create telecom’s first truly autonomous workforce by 2030
Breadth substitutes for intimacy
Personalisation is usually assumed to need more personal data. It does not have to. A system that sees wide behavioral signal across telecom, payments, content and gaming can personalise on patterns rather than identity. The more it sees of what people like you do, the less it needs to know about who you are.
Where trust is actually earned
Most subscribers do not decide whether to trust an operator by reading its privacy policy. They decide in small, frequent exchanges: whether points are credited when promised, whether a voucher works at the till, whether an offer feels useful or intrusive.
A user who watches food content and sees grocery vouchers surface first experiences AI as a service. The same signal used to push products they never asked for reads as surveillance. Every honored reward is a deposit. Every offer that feels like tracking is a withdrawal.
That exchange matters beyond churn. The GSMA estimates that 3.1 billion people live within mobile broadband coverage but do not use it. For many of them, the question is not whether the network reaches them. It is whether being online is safe enough to be worth it.
Operators planning AI launches for 2027 should be able to answer two questions before signing a vendor. Can our own security team test this before it goes live? Can we switch it off tomorrow? If the answer to either is no, the AI is not ready, however good the demo looks.
—
Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.
The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.
Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.
The post Telcos are racing to put AI in the app. Their subscribers have already said which kind they will trust appeared first on e27.