Staff in many offices already use AI chat assistants to draft letters and shorten long reports, often on their own phones and without any office rule. The tools are useful, and they are also confidently wrong often enough to cause real harm. A public office needs a clear idea of which tasks belong to the tool and which stay with a person.
What a large language model is
The assistants most people know are built on a large language model. This is a computer program trained on a very large amount of text. The United States National Institute of Standards and Technology, known as NIST, explains the core of how it works: the model predicts the next word, or piece of a word, in a sentence, and its output follows the statistical patterns of the text it was trained on.1
This one fact explains both the strengths and the weaknesses. A program that is good at predicting the next word writes smooth, well-organized text in many styles and languages. It does not look up a fact and then report it. NIST notes that this kind of prediction can produce accurate answers, and can also produce answers that are factually wrong or that contradict themselves.1
Where it helps
NIST lists the common uses of these systems as text generation and editing, summarization, search and chat, along with writing computer code.1 In an office, that points to work where a person already knows the subject and can check the result quickly.
We think three uses are worth trying first. One is a first draft of a routine letter, notice or minutes, which the writer then corrects. Another is a summary of a long document the reader also has on hand. The third is finding the right passage in a large set of the office's own files, where the assistant shows the page it took the answer from.
In each case the person stays the author. The assistant saves typing time. It does not replace knowing the subject, and the name on the document is still the name of the person who signed it.
Where it fails
The first failure is made-up content. NIST calls this confabulation: the system produces false content and presents it with confidence. It can also invent the reasoning or the citations that seem to support its answer. NIST describes this as a natural result of how these models are designed, and notes that made-up legal content has been shown to be widespread in current models.1
For an office, the lesson is practical. A draft memo that cites a section of a law, a resolution number or a budget figure needs every one of those checked against the real document.
The second failure is in the reader. NIST warns of automation bias, which is giving too much trust to an automated system, and says people may come to rate machine-written content higher than it deserves.1 A tidy paragraph looks finished, and that is exactly when a tired reviewer stops checking.
The third is private data. NIST notes that these models may leak, generate or correctly infer sensitive information about individuals.1 In the Philippines, the Data Privacy Act of 2012 applies whenever personal information is processed, and the office that controls a record stays responsible for it even after the record is passed to a third party for processing.2 Typing a resident's name, health condition or case details into a public chat tool hands that record to a company outside the office, so our advice is not to do it unless the office has an approved tool and a lawful reason.
Decisions stay with people
International guidance agrees on one point. The UNESCO Recommendation on the Ethics of Artificial Intelligence, adopted by 193 member states in November 2021, lists human oversight and determination among its principles and says AI systems should not displace ultimate human responsibility and accountability.3 The AI principles of the Organisation for Economic Co-operation and Development, adopted in 2019 and updated in May 2024, include transparency and explainability, and accountability.4
Philippine rules already give people a say when a machine is involved. Under the rules of the Data Privacy Act, a person has the right to be informed that automated decision-making or profiling exists, and the right to object to automated processing or profiling.5
So a decision that affects a person's rights, benefits, job or record belongs to an accountable officer. That covers who receives assistance, who is hired, and whether a permit is granted. An assistant may help prepare the papers. A person reads them, decides, signs, and can explain the reason afterward.
The Philippine government's own rules
On June 9, 2026, the Department of Information and Communications Technology and the Civil Service Commission issued Joint Memorandum Circular No. 003, series of 2026, on the ethical and responsible development, deployment and use of AI in the government. It applies to all government officials and employees in constitutional bodies, national agencies, government corporations with original charters, local government units and state universities and colleges.6
The circular states that AI systems shall not displace ultimate human responsibility and accountability, and that an agency remains accountable when it outsources the work. AI users affected by an AI decision have the right to contest it and ask the agency for a human review. An office that deploys AI must disclose to the public or the stakeholders concerned whether AI models and systems were used in its deliverables and services.6
Each agency is to form an internal committee with at least its information technology and legal units to advise the head of agency, and must handle government data in AI systems under the Data Privacy Act and related rules. The two agencies are to issue implementing guidelines within 90 days of the circular taking effect.6 Offices in the Bangsamoro should read the full text and confirm with their own legal and IT staff how it applies to them.
What you can do this week
An office does not need to buy anything to start using these tools with care. We suggest agreeing on a one-page rule and reading it out at the next staff meeting.
The rule can be as short as the six lines below. What matters is that every staff member has heard it.
- Ask staff which AI tools they already use for work, without blame, so the office knows where it stands.
- Agree that no personal information about residents, clients, patients, students or staff goes into a public AI tool.
- Agree that every name, date, figure and legal citation in an AI-assisted draft is checked against the source document before signing.
- Keep decisions about a person's rights or benefits with a named officer.
- Say so on the document when AI helped prepare it.
- Ask your legal and IT staff for the text of the 2026 joint circular and any implementing guidelines.
Sources
- Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1). National Institute of Standards and Technology (NIST), 2024-07.
- Republic Act No. 10173 (Data Privacy Act of 2012). The LawPhil Project, Arellano Law Foundation, 2012-08-15.
- Recommendation on the Ethics of Artificial Intelligence. UNESCO, 2021-11.
- OECD AI Principles. Organisation for Economic Co-operation and Development (OECD), 2019, updated 2024-05.
- Implementing Rules and Regulations of the Data Privacy Act of 2012. National Privacy Commission, 2016-08-24.
- Joint Memorandum Circular No. 003, s. 2026: Principles for the Ethical and Responsible Development, Deployment, and Use of Artificial Intelligence (AI) in the Government. Department of Information and Communications Technology and Civil Service Commission, copy on the Official Gazette, 2026-06-09.