VIỆT NAM'S BUSINESSES MOVE TOWARDS AI AGENTS

Banking and financial services, telecommunications and IT are among the sectors leading AI agent trials, supported by their existing digital infrastructure.

HÀ NỘI — Việt Nam’s AI landscape is entering a new phase as businesses move beyond generative AI (GenAI) tools designed to assist human users towards autonomous AI, agentic AI and AI agents that can take action with limited supervision, experts have said.

The shift comes as Việt Nam seeks to accelerate science and technology, innovation and digital transformation under Politburo Resolution 57-NQ/TW, issued on December 22, 2024.

Regional Technical Head at ManageEngine Phạm Huân said the key difference between basic GenAI and autonomous AI agents lies in their degree of autonomy and ability to execute tasks.

GenAI typically works through a prompt-and-response model, generating content such as text, images, video and audio in response to user requests. Autonomous AI agents, by contrast, are designed to make decisions and take actions independently in pursuit of more complex goals.

Huân said that in IT operations, for example, GenAI can guide an employee through a troubleshooting procedure, while an autonomous AI agent can detect an issue, analyse its root cause and execute a system recovery script.

GenAI suggests what you should do, while an AI agent actually gets on with completing the task.

"Việt Nam is emerging as one of Southeast Asia’s most dynamic AI adoption markets, ranking second in the region with 26.5 per cent of its working-age population using AI, while also ranking among the world's top 15 economies for AI user-growth speed," he said.

However, Vietnamese enterprises remain at an early stage of deploying autonomous AI agents. Many are transitioning from GenAI to autonomous AI, with applications still largely focused on customer support chatbots, document automation and other basic use cases.

Banking and financial services, telecommunications and IT are among the sectors leading AI agent trials, supported by their existing digital infrastructure.

He said the main barriers to wider adoption are weaknesses in data governance, limited AI transparency and cybersecurity risks associated with giving AI tools greater access to core data. Once data standardisation and security challenges are addressed, AI agents could become an important driver of greater operational efficient and lower costs.

Greater autonomy also creates new cybersecurity risks, however. One of the most significant is indirect prompt injection. Instead of attacking an AI system through prompt entered directly by a user, attackers can hide malicious instructions in documents, emails or PDF files that an AI agent processes during its normal workflow.

Employees may also independently connect external AI agents to corporate data without IT teams' knowledge, creating data flows that may not be properly monitored or audited.

He said businesses should therefore adopt a defence-in-depth approach. AI agents should not receive unrestricted privileges, while identity and access controls should be applied throughout their operation. Humans should retain final approval over critical decisions and actions, with an emergency stop mechanism available when necessary.

For SMEs, an effective AI governance framework does not have to mean lengthy policy documents. Businesses can classify AI applications according to risk, allowing relatively unrestricted use of AI for non-sensitive creative tasks while requiring human oversight for internal data and customer interactions. The highest level of control should apply to AI systems handling core or financial databases.

The principle of least privilege should also be mandatory, with AI systems granted access to only the data required for their specific tasks. Enterprises should maintain human-in-the-loop checkpoints for critical operations and use centralised IT management and cybersecurity tools to monitor AI applications and data flows.

He said successful deployment is ultimately not about having the smartest AI agent, but about establishing a robust security framework in which AI can optimise processes within controlled boundaries.

For Việt Nam, the challenge will be to capture the productivity benefits of increasingly autonomous AI while strengthening the data governance, security and human oversight needed to support its wider adoption. — VNS

Source: VNS