More businesses and consumers are turning to artificial intelligence (AI) agents to make decisions, but who exactly do they blame when something goes wrong?
As it is, 75% of Southeast Asian consumers are willing to let AI agents autonomously find the best prices and complete purchases on their behalf.
Some 72% already use everyday automations, such as price alerts and automatic payments, according to this year’s iteration of the e-Conomy SEA report from Google, Temasek, and Bain & Company. The study analysed 10 countries in the region, including Singapore, Indonesia, Thailand, and the Philippines.
In fact, a potential increase in agentic adoption could fuel a 60-fold growth in AI adoption across the region by 2030, the report noted.
Asked what would make them more willing to let AI act on their behalf, 44% of consumers pointed to strong privacy and data protection safeguards.
Another 36% highlighted clear information about what AI could or could not do, while 30% cited the ability to reverse or undo AI actions, the study found, polling 9,006 respondents.
About 45% of AI users were open to delegating purchase decisions to agent within specifications and guardrails, while 50% would allow AI agents to make transactions exceeding US$10.
“Consumers are receptive to the idea of autonomous assistance, but remain cautious about unknown variables,” the report noted. “Adoption can be further accelerated if platforms provide clear in-app visibility into capabilities and are operated by a trusted company which products they already use.”
Accountability must accompany agentic transactions
For one, reverse liability is essential as agentic AI adoption picks up momentum, said Nitin Gajria, Google Asia-Pacific’s Southeast Asia and South Asia frontier markets managing director.
This need is no different from online commerce, where customers want the ability to return their orders if their purchases are unsuitable, such as when a coat is too large, Gajria said, during a media briefing to discuss the study findings.
He pointed to key conditions that are critical to sustain agentic AI growth, including the availability of agentic applications and trusted infrastructures on which such services are delivered.
Such trusted ecosystems must have safeguards such as data privacy protection and reversibility, in which agentic actions can be undone, he said.
These also are necessary measures to mitigate security risks, such as rogue AI agents and prompt injection that can result in unauthorised agentic actions.
Agentic commerce can only take off if it is conducted in a safe environment and with a high degree of transparency, where users understand why an agent has made the decision it made, Gajria said.
And they must have the ability to reverse a decision an agent makes, he said.
There currently also is a lot of room for the right legal frameworks and governance to evolve, said Florian Hoppe, partner at Bain & Company.
For one, while consumer digital IDs are well established and maturing, there still is a gap in agentic identity management, Hoppe said.
For agentic transactions to scale, a “Know Your Agent” layer needs to be integrated over existing rails, he said.
One such industry effort has come from the Monetary Authority of Singapore, which in July revealed its Safeguards for AI Agents in Finance (SAFR) at Runtime initiative, according to the e-Conomy report.
The Singapore framework proposes guidelines that will enable AI agents in financial services to carry out financial tasks “safely, securely, and reliably”.
It provides “governance checkpoints” that verify and record an AI agent’s proposed actions before these are executed, ensuring they act within predefined mandates and policies.
Underscoring the need for accountability and governance, Hoppe said there has to be clear rules and understanding on whether a faulty purchase can be reversed and, if not, who is at fault when something goes wrong — whether it is the AI agent or user.
Strong legal frameworks are necessary for consumer protection, including in agentic commerce, he added.
“We have to be clear on what the rules are on this,” he said.
Coordinated efforts needed in frontier AI
There also needs to be more clarity on how foundation AI models are being advanced and the potential impact they have on organisations and consumers.
This is especially apparent in light of recent developments involving rogue AI agents, in particular, AI agents powered by OpenAI that staged a July hacking campaign against Hugging Face.
There has to be more thoughtful development on frontier models and coordinated industry effort on governance, said Gajria, who noted that this now is starting to take place in a more coordinated manner.
Such efforts signal a positive step forward, but organisations and consumers still should practise caution, he said, in response to FutureCIO’s question on how the recent developments would impact agentic AI adoption.
Google, OpenAI, and Anthropic reportedly are working to create a unit, temporarily coined Standards Authority for Fronter AI (SAFA), that will set guidelines on risk assessment and testing for frontier models. The entity will operate independently of government control and is said to launch early-2027.
Meanwhile, generative AI (GenAI) adoption in Southeast Asia has hit 243 million people, or 58% of the working population, according to the e-Conomy SEA report.
Quantifying AI usage in tokens, the report further assessed that consumers account for 60% of the region’s tokens, tapping them mostly for work and study tasks and leaving enterprise use with significant headroom to scale.
Enterprise workloads are strongest in more mature markets, driving 45% to 60% of use in Singapore and Malaysia, the report noted.
Forrester also predicts that enterprise buyers will increasingly deploy AI agents to handle tasks across the buying process, including finding suppliers and analysing contract renewals.
This will push providers to ensure they can respond with speed, accuracy, and consistency, the research firm said.
Interestingly, it will fuel more human interactions as buyers look to validate AI-generated recommendations and automated outreach.
Since AI cannot fully discern service provider fit, trust, or accountability, buyers will double their conversations with providers and trusted third parties, such as customer references and partners, to confirm their decisions, Forrester said.
In addition, the research firm noted that two-thirds of providers will lose out on deals when buyers cannot verify claims.
“As AI becomes ubiquitous across the buying process, B2B leaders that pair AI-driven speed and convenience with transparency, control, and trust will extract the most value from AI,” said Sharyn Leaver, Forrester’s chief research officer.
