AI code spends money on its own, so how do banks keep control?

This article summarizes an episode of Exponential’s video series featuring Pei Ling Tin, co-president of MetaComp.

Pei Ling Tin, co-president of MetaComp/ Photo credit: Pei Ling Tin

Traditional banking websites are becoming irrelevant. Pei Ling Tin, co-president of MetaComp, says financial services are shifting into chat platforms like WhatsApp and Telegram.

Powered by agentic AI, these tools let users deploy software agents to handle transactions, removing a bank’s control over the customer interface.

When software executes trades and signs contracts autonomously, traditional compliance models become a liability. Firms across Southeast Asia must build technical boundaries, pivot to partner-driven sales, and rethink training to solve talent shortages.

The end of the traditional banking website

Redesigning how firms handle sales starts with abandoning the centralized website. Banks face a structural threat as financial services migrate directly into messaging apps, removing an institution’s ability to control the user journey.

“With agentic AI, you no longer have to navigate multi-step portals,” Tin explains. “You don’t have to go to where the service provider is. From where you are, from your phone, your prompt, your WhatsApp, and Telegram, you can start to access these services.”

Government rules slow down new features
Pushing these conversational tools to market collides with existing compliance laws. While the customer experience can be transformed, legal requirements cannot be compromised, she cautions, noting that agentic AI will inevitably define what comes next.

The risk of letting code make business decisions

Beyond regulatory delays, pushing AI into messaging apps introduces operational risk. Financial firms must now manage autonomous software that makes independent choices rather than executing basic, hard-coded commands.

“Agents can autonomously act on our behalf, whether to negotiate contracts or to decide and make decisions on very material matters, such as making a transaction,” Tin argues.

To manage software acting independently, financial leaders must build technical kill switches. After determining who is deploying the tool, the next step is defining what the system can do, establishing clear thresholds past which it cannot proceed or overreach.

Creating rules to hold machines accountable

Establishing these technical boundaries requires regulators to track software programs throughout their entire lifecycle. To survive this shift, firms must build limits that define a new standard for government oversight across these areas:

• Agent identity and registration links each AI program to a human owner to verify who stands behind automated actions.
• Authority and permission defines strict limits that prevent programs from acting beyond approved activities.
• Behavioral monitoring applies dynamic risk scores based on the AI’s ongoing actions.
• Ecosystem governance mandates rules for software interaction and ensures data travels with digital actions.
• Governing principles enforce these limits by tying user responsibility directly to AI ownership.

Old compliance plans fail modern code
Traditional rules cannot govern autonomous machines. “Traditional financial frameworks, whether it’s KYC, AML, or CFT, are static, and they’re meant for the human world,” Tin notes. “When agents are machines, we have to have frameworks that treat them as identifiable and accountable.”

Relying on partners to grow payment networks

Losing workers to caregiving threatens AI’s financial benefits