ANALYSIS: The BIR is testing AI for tax officers. What happens when AI enters decision-making?

The BIR’s new tax AI assistant could change how government workers find, verify and apply tax rules, while raising new questions about human judgment, data security and accountability.

The Bureau of Internal Revenue’s new AI project with Ayala Corp. offers an early look at what happens when artificial intelligence becomes part of the way government workers do their jobs. From finding the right rules to helping them arrive at decisions that can affect taxpayers and businesses, AI could become part of work that once relied almost entirely on human research and judgment.

Under a six-month partnership, Ayala will help the BIR set up “Terry the Tax AI Assistant,” a tool designed to help BIR personnel retrieve, cross-reference and analyze tax regulations, rulings, issuances and other official tax materials. The project will be provided at no cost to the government, with BIR personnel taking part in its testing and refinement.

That setup is important because the BIR is not simply handing its employees an AI tool and asking them to use it. Its own people will help test Terry, check its responses and refine the system based on the bureau’s actual needs.

That could become one of the most important parts of the experiment.

AI at work is often framed around automation and how many tasks technology can take over. But tax administration involves rules that can have real consequences for people and companies. Terry may be able to find a relevant regulation in seconds, but someone still has to determine whether that rule applies to a particular case.

BIR Commissioner Charlito Martin R. Mendoza said the value of AI should not simply be about getting answers faster. It should help employees find and verify the right authorities and make “sound and defensible decisions.”

That approach could eventually be felt by taxpayers and businesses dealing with the BIR. If officers can find the relevant rules more quickly and apply them more consistently, it could mean faster responses and less time spent going back and forth over questions involving complicated tax regulations.

But putting AI into the process also creates a new responsibility for the people using it.

The faster a tool produces an answer, the easier it can be to accept that answer without checking where it came from. That is why Terry’s ability to provide source-grounded responses using official BIR references is important. The technology is being designed to help employees find and verify the basis for an answer, rather than simply generate one.

There is also the question of data privacy and security. As government agencies bring AI deeper into their operations, they will have to be careful about what information goes into these systems, who can access it and how it is protected. The BIR handles information that can be highly sensitive to taxpayers and businesses, making those safeguards especially important as the technology expands beyond searching tax rules and official issuances.

The BIR’s plan to involve its personnel in testing and refinement is also significant. The partnership includes knowledge transfer to help the bureau institutionalize Terry and continue developing it as its needs evolve.

In effect, the BIR is not only building an AI assistant. It is also building a workforce that needs to know how to work with one.

That lesson extends beyond government. Businesses adopting AI face a similar challenge. Buying the technology is one thing. Teaching employees when to use it, when to verify its output and when human judgment needs to take over is where much of the real work begins.

The BIR’s experiment could therefore be judged by more than how well Terry performs after six months.

If it helps BIR employees work faster while making them better at finding, checking and applying tax rules, the project could offer a useful model for how AI enters the workplace without taking human judgment out of it.

Because when AI becomes part of decision-making, the bigger change may not be who does the work. It may be how the people doing it learn to work with the machine.

READ: