Asia-Pacific organisations already are applying agentic artificial intelligence (AI) in their workflows, but are selective about where the AI agents are deployed.
The move to hold back may be warranted, especially since not all organisations are able to fully account for their AI agents’ decisions, according to Sumsub’s Asia-Pacific State of Digital Trust report, released on Tuesday.
It revealed that just 50% of organisations in the region are able to reconstruct a decision pathway and only 38% can run a tamper-proof audit trail.
This is despite 95% of respondents expressing confidence they can explain an AI decision, the study found.
The research attempts to assess the “readiness of autonomy, responsibility, and traceability” in AI, drawing responses from 720 respondents across nine Asia-Pacific markets, including Singapore, Australia, India, China, and Thailand.
Conducted between April and June 2026, the survey was carried out by Blackbox Research, in partnership with the Singapore Fintech Association. Respondents comprised business decision-makers with responsibility for AI strategy and oversight, and from four sectors: financial services, IT and software services, e-commerce, and mobility and delivery platforms.
The report focuses on the three areas — autonomy, responsibility, and traceability — to evaluate how effectively the respondents are governing their AI agents.
Autonomy looks at the extent AI is acting independently, while responsibility looks at how clearly the ownership of AI-driven outcomes is defined and traceability assesses how well these decisions are reconstructed and explained.
The region’s autonomy score is 69.9%, while responsibility is 70.3% and traceability tips at 61%.
India tops the pack in autonomy with a score of 71.8%, while Thailand and the Philippines tie for lead in responsibility at 74.8%, and Thailand has the highest traceability score of 64.8%.
Region-wide, by sectors, financial services lead in responsibility and traceability at 72.4% and 63.3%, respectively, while IT and software services lead in autonomy at 73.7%, just slightly higher than the financial services’ score of 73.1%.
In addition, 68% of financial services organisations maintain audit trails of AI decisions, clocking the highest across the verticals. However, just 41% hold tamper-proof records.
Plug the gaps or face the risks
Organisations are falling behind in terms of traceability, said Penny Chai, Asia-Pacific vice president at Sumbsub, which provides AI verification solutions.
As AI agents gain more autonomy, there will be blindspots that need to be resolved, Chai said at a media briefing to discuss the study findings.
If these are not traceable, they cannot be corrected and will, hence, create risks to organisations, she said.
In Singapore, 94% are using or piloting agentic AI, but just 29% can produce the evidence to verify the AI decisions, the study found.
And while 90% of Singapore respondents have strong comfort levels in letting AI handle low-risk routine tasks, they observe more prudence when financial liabilities are applied.
Some 16% have significantly increased the scope of autonomy of their AI systems over the past year.
About 40% have explicit guidelines that assign direct responsibility to a specific person for AI outcomes, while 30% direct that responsibility to a team.
Some 29% in Singapore also view AI to have the greatest real-world impact on data-related tasks, while 21% point to operations or workflow processing and 15% cite the technology’s impact on critical security applications, such as fraud detection.
“Everyone is focused on how quickly AI is advancing, but the bigger question is whether governance is keeping pace,” said Holly Fang, president of Singapore Fintech Association.
She noted that fewer than one in three organisations in Singapore are able to produce an audit trail for AI-driven decisions.
“As AI moves beyond copilots into autonomous agents handling increasingly critical workflows, the focus now should be on building the traceability, accountability, and governance needed to deploy AI at scale,” Fang said. “As an industry, that’s where our attention needs to be next.”
Across Asia-Pacific, 68% say AI has taken on autonomous roles in operations and workflow processing, while 63% have applied the same to customer service and support.
Another 53% have AI take on autonomous roles in internal decision-making and analytics, and 42% have done likewise for software and engineering workflows.
Some 39% let AI take on autonomous roles in fraud detection, anti-money laundering, and risks monitoring, while 35% do so for compliance and case review, and 20% have done likewise for payments and financial transactions.
Asked what they are most concerned about when agents act on their behalf, 52% of respondents point to incorrect or unreliable decisions. Another 44% cite regulatory or compliance risk, while 43% highlight data quality of input issues.
Some 33% are worried about lack of transparency or explainability, and 31% cite lack of human oversight.
Only 31% have AI initiating or completing multi-step tasks with limited human input, the study found.
Trust comes first before scale
The findings suggest organisations are choosing to be cautious when it comes to payments and financial transactions, than in other business functions, Nimit Gulati, vice president of acceptance solutions, value-added services at Visa Asia-Pacific, said in the report.
It reflects a higher bar for accountability when money is involved, Gulati said.
The real challenge is not whether AI can make decisions, but whether the decisions it makes can be trusted, he said.
The findings further highlight the importance of transparency, security, and accountability in AI-driven commerce, he added.
Fang noted that the journey towards agentic AI likely will be incremental, with fintechs adopting AI agents first in lower-risk workflows, while retaining human oversight for decisions that involve customer onboarding, movement of client funds, and other regulated activities.
“Building confidence in governance will be just as important as advances in the technology itself,” she said.
Chai added that rather than a lack of strategic intent, organisations also are opting for prudence in terms of scaling their AI agents.
“When financial liabilities are on the line, immature traceability systems create an unacceptable operational risk,” she said.
She urged the need for organisations to establish tracking architectures and guardrails to ensure they can safely deploy “high-stakes AI” at scale.
According to the study, more than nine in 10 respondents acknowledge stretching an AI system beyond its original purpose over the past year, with six in 10 already having experienced an autonomous AI action gone wrong.
In fact, failed pilots can affect users’ willingness to use AI in their work.
Just 29% of Singapore desk workers describe themselves as AI sceptics, but only 6% say AI is a core part of their daily work, revealed a study by Salesforce last month.
Globally, 37% of desk workers say they are AI sceptics, while 11% describe AI as a core part of their daily work. The online survey polled more than 1,500 respondents across 14 markets, including India, Japan, Germany, Italy, and the UK.
The Salesforce report attributed the findings to disappointment over AI rollouts, with 31% of Singapore respondents having experienced unsuccessful AI pilots. Some 38% express low trust in outputs, higher than the global average of 28%.
Another July 2026 study by Boomi also found that while 86% of organisations have deployed AI agents, a much lower 34% trust the actions taken by these agents.
Conducted by Forrester Consulting, the study surveyed 409 IT decision makers in Asia-Pacific, Europe, and North America.
Amongst organisations that have agentic control or readiness, 55% have high confidence in their agents’ actions and decisions, compared to 22% of those in “agentic chaos”. The latter is tagged to organisations that are in the lower end of operational readiness in various areas, including governance, integration, and API and MCP management.
Boomi’s chairman and CEO Steve Lucas noted that agents should only be trusted to act on data that has been properly connected and governed.
Singapore in January launched its Model AI Governance Framework for Agentic AI, designed to guide organisations on how to deploy AI agents responsibly. It includes technical and non-technical measures to mitigate risks.
