CyberCode.ph · Philippines

Business AI Compliance in the Philippines: Governance and Approval Checklist

Last updated October 3, 2026 · Practical privacy, cybersecurity and technology-law guidance
AI governance and business technology interface for Philippine companies

Last materially reviewed: September 6, 2026

Direct Answer

Business AI compliance starts with the use case: what data enters the tool, what it does and who may be affected. Philippine privacy duties apply when AI processes personal data; contracts, intellectual property, employment and sector rules may also matter. This guide helps you document an AI use case, review the vendor and assign human oversight. For the dated status of legislation and issuances, see the Philippine AI law tracker.

This implementation guide is part of Cybercode’s broader AI governance and emerging-technology hub.

Business AI Compliance: Start With One Use Case

Assess the task, data and consequences of an AI tool before approving it. Use the questions below with your business owner, privacy lead and IT team. Record unresolved issues as “unknown”; a missing answer is a reason to investigate, not evidence that a tool is safe.

Assess an AI tool · Copy the approval record · See a worked example

Can We Use This AI Tool?

This is a recommended internal review workflow, not a legal certification or a government risk rating. Open each question and record the evidence supporting your answer.

1. What task will the tool perform, and who owns it?

Describe one use case: drafting replies, summarising meetings, screening applicants or taking actions in company systems. Name the accountable owner, users and permitted actions. Start with synthetic test data while unresolved data or access issues are reviewed. Include AI features inside software you already use.

2. Will personal, confidential or client data enter the tool?

Map prompts, uploads, recordings, connected accounts and logs. For personal data, establish the applicable lawful basis and privacy safeguards before processing. Review NPC AI privacy guidance. If staff have already uploaded customer data, follow the response guide. Removing a name alone may not make data anonymous.

3. Will the output score, profile or significantly affect a person?

Check hiring, discipline, credit and similar decisions separately from ordinary drafting. Review registration, notification and human-review requirements. NPC Circular 2022-04, Section 5(A), covers personal-data processing systems involving automated decision-making or profiling. Do not assume a small business is exempt.

4. What does the vendor do with inputs and outputs?

Obtain written answers on model training, retention, deletion, subprocessors, locations, security and incidents. Compare the contract with the settings actually enabled. Use the AI vendor checklist and input and output contract guide.

5. Who checks the output and can stop the tool?

Define an approver for external communications and consequential actions. Test errors, inappropriate disclosures and tool permissions. Keep a practical way to suspend access. Where automated decisions pose significant risks to data subjects, NPC guidance calls for meaningful human intervention and a way to contest the decision. Use the AI security guide.

6. What must be documented before staff begin?

Keep the completed record below, relevant vendor evidence, test results and any required privacy assessment. Assign actions and deadlines for unresolved issues. Adapt the AI usage policy template to the approved tools and data restrictions. Reassess material changes to the vendor, data, permissions or purpose.

Suggested internal decision: “Hold for review” when a material issue is unresolved; “limited pilot” when a defined test can proceed with documented controls; or “approved for this use case” after the responsible people complete their review. Approval for drafting does not automatically authorize profiling, customer-data uploads or autonomous actions.

Copyable AI Inventory and Approval Record

Copy this list into your organisation’s own document. Keep confidential details in your internal records. This is a working template, not an NPC filing form.

  • Use-case ID, date and business owner: [complete]
  • Tool, vendor, plan/version and purpose: [complete]
  • Approved users, connected systems and actions: [complete]
  • Inputs, outputs and affected people: [complete]
  • Personal/confidential data and permitted restrictions: [complete]
  • Lawful-basis, notice, PIA and registration review: [finding, evidence or unresolved question]
  • Training, retention, deletion and transfer terms: [document links and settings checked]
  • Human review, access controls and stop procedure: [complete]
  • Test results, remaining issues, owner and due date: [complete]
  • Decision, conditions, approver and review trigger: [complete]

After a use case is approved: transfer its restrictions into the staff rollout pack and train authorised users before granting access.

Worked Example: A Retailer Drafting Customer Replies

Fictional example: a Philippine retailer wants an AI assistant to draft replies from customer emails. The proposed connection would expose names, addresses and order histories. The team records the purpose and data flow, requests vendor terms and asks its privacy lead to assess the intended processing.

Illustrative pilot decision: test with invented customer messages first; allow no inbox connection, live customer data, automatic sending or refund actions. A staff member checks each draft. Before any live-data rollout, resolve the privacy assessment, vendor terms, access restrictions and applicable registration questions. Keep the test results and approval conditions with the inventory record.

Your next step: complete one record, then use the complete worked AI privacy impact assessment, vendor review and staff policy template to resolve the gaps it reveals.

Targeted toolkit review: 3 October 2026. Legal anchors: NPC Advisory 2024-04, Sections 1–3 and NPC Circular 2022-04, Section 5(A). The record and pilot workflow are editorial recommendations. The historical status discussion below retains its stated review date.

Key Takeaways

  • There is no Philippine AI Act. Bills have been filed; none is in force as of September 6, 2026.
  • NPC Advisory No. 2024-04 binds you where an AI system processes personal data — which covers most business AI in practice.
  • Section 2(B) requires appropriate and effective governance mechanisms. Its examples include PIAs, privacy-by-design, monitoring and an AI ethics board; the list is not a requirement that every business establish an ethics board.
  • Where automated decisions pose significant risk to rights and freedoms, human intervention must be by “persons with the necessary competence and authority,” and data subjects must be able to contest.
  • Accountability cannot be outsourced: controllers “shall remain accountable for the actions of their PIPs.”
  • NPC Circular No. 2022-04, Section 5(A), requires registration of systems processing personal or sensitive personal information involving automated decision-making or profiling.

Jump to a Section

What Actually Binds You Today

Most published guidance on Philippine AI governance describes aspirations rather than obligations. This table separates them. Status as of September 6, 2026.

Instrument Issuer Date Who it binds What it requires
RA 10173 and its IRR Congress; NPC 2012; IRR as amended Anyone processing personal data Lawful basis, privacy principles, data subject rights, security, accountability, DPO, records of processing
NPC Advisory No. 2024-04 NPC 19 Dec 2024 PICs and PIPs where personal data is processed in AI development or deployment, including training and testing Transparency on AI factors and inputs; accountability; governance mechanisms; human intervention; fairness; accuracy; minimisation
NPC Circular No. 2022-04 NPC 5 Dec 2022 Covered PICs and PIPs; in all instances, systems involving automated decision-making or profiling Registration, ADM/profiling notification, DPO designation and registration
NPC Circular No. 2023-06 NPC 1 Dec 2023 PICs and PIPs, public and private Security measures; PIA; privacy-by-design; access control; logging; retention and disposal
NPC Advisory No. 2017-03 NPC 31 Jul 2017 PICs and PIPs PIA before adoption of new systems, unless risks are minimal
NPC Advisory No. 2026-01 NPC 13 Apr 2026 Those who scrape, and PICs hosting publicly available personal data Purpose specification, lawful basis, notice, minimisation, mandatory PIA
Sector issuances (e.g. BSP, Insurance Commission) Sector regulators Varies Only entities that regulator supervises Sector technology, cyber and risk-management requirements
National AI governance framework DEPDev, under NAIS-PH Reported still in development, 8 Aug 2026 Not binding on private business Nothing enforceable yet

Governing Law and Authorities

The operative authority is RA 10173 and its Implementing Rules and Regulations. The National Privacy Commission enforces them, with power under Section 7 of the Act to investigate, issue orders and publish guidance.

Advisory No. 2024-04 creates no new AI regime. Its premise is that the DPA, its IRR and NPC issuances are “principles-based and technology-neutral,” so they apply “when the processing of personal data is involved in the development or deployment of AI systems, including its training and testing” (Section 1). Where an AI use case genuinely touches no personal data, the Advisory does not reach it, and your duties come instead from contract, confidentiality, intellectual property and employment law. Few enterprise deployments stay clear of personal data for long, so build one framework and apply it to everything.

The Obligations, Element by Element

Inventory and ownership

Philippine law names no “AI inventory,” but it underpins two things that are required: the records of processing activities under IRR Section 26(c), which must “sufficiently describe” the system and identify who has access, and registration under NPC Circular No. 2022-04. Record tool, purpose, owner, inputs, outputs, whether the system can act, and whether a person is affected. Include employee-introduced tools — accountability attaches to the controller regardless of who started using them.

Risk classification

Advisory No. 2024-04 imposes no tiered risk taxonomy, and you should not claim it does. It sets a threshold that works as a classifier: where automated decision-making poses “a significant risk to the rights and freedoms of data subjects,” specific safeguards become mandatory. Build tiers around that language, not a foreign scale. Treat as higher risk anything that scores, screens, prices, disciplines, monitors or excludes a person, touches sensitive personal information, or acts without review.

Human oversight

This is the Advisory’s sharpest requirement. Section 2(B) requires “mechanisms to allow for meaningful human intervention carried out by persons with the necessary competence and authority,” and separately requires mechanisms letting data subjects “question and contest automated decisions.” Competence means enough understanding to disagree with the system; authority means standing to overturn it without an escalation that never happens. The IRR adds a harder rule: under Section 48(b), “No decision with legal effects concerning a data subject shall be made solely on the basis of automated processing without the consent of the data subject.” Our guide to automated decision-making and profiling works through that provision.

The mechanisms the Advisory names

Section 2(B) lists what the NPC expects instituted: Privacy Impact Assessments; privacy-by-design and privacy-by-default; common industry security standards; continuous monitoring of AI operations; a dedicated AI ethics board; regular retraining and scrubbing; and human intervention in decision-making and output review. The same section makes the controller “responsible for regularly monitoring the proper implementation and effectiveness of these governance mechanisms” — the framework must be audited, not merely written.

Fairness, accuracy and minimisation

Section 2(C) requires processing that is “neither manipulative nor unduly oppressive,” with mechanisms to identify, monitor and limit bias, expressly covering “systemic bias, human bias, and statistical bias.” It also prohibits AI Washing — overstating AI involvement “to the detriment of their data subjects.” Section 2(D) requires accuracy of personal data so output stays fair. Section 2(E) requires exclusion by default of “any personal data that is unlikely to improve the development or deployment of AI systems.”

Roles and accountability

IRR Section 26(a) requires designation of a data protection officer or compliance officer; Circular No. 2022-04 Section 8 requires registration to be filed through that DPO and permits only one registered DPO. Advisory No. 2024-04 fixes accountability on “the specific PIC undertaking the development or deployment of the AI system,” and confirms controllers “shall remain accountable for the actions of their PIPs.” Name a system owner per use case, a DPO with visibility across all of them, and an approver for higher-risk deployments.

Vendor management

Most Philippine businesses buy models rather than build them, so vendor governance is the operating core. The Advisory requires controllers and their processors to “demonstrate that they have implemented effective policies and procedures to comply with the DPA.” For cross-border processing, NPC Advisory No. 2024-01 (30 May 2024) offers model contractual clauses but is explicit that use is voluntary — “The NPC does not require contractual parties to adopt the use of MCCs” — and that it “will not accept any requests for review of agreements or contracts” for conformity. Model clauses are not NPC approval. See our AI vendor due diligence checklist.

Monitoring, incidents and review

NPC Circular No. 2023-06 supplies the security floor: need-to-know access control with authentication (Section 13), multifactor or similar secure authentication for online access to sensitive data (Section 16), longer retention of logs on authentication attempts and security incidents (Section 29), documented retention periods (Section 9) and secure disposal (Section 30). Advisory No. 2024-04 adds continuous monitoring of AI operations.

What Is Not Settled or Not Yet Issued

Honest governance depends on knowing where the rules stop. As of September 6, 2026:

  • No Philippine AI statute exists. Bills addressing artificial intelligence have been filed in Congress. A filed bill is not law, confers no obligations, and must never be presented to management as a compliance deadline.
  • No national AI governance framework binds private business. IPOPHL reporting dated 8 August 2026 describes DEPDev as lead agency for AI and data policy under the National AI Strategy for the Philippines, working to develop “the first national AI governance framework” — that is, still in development. Treat any document circulating as a “National Responsible AI Governance Framework” or a “Framework for Ethical and Responsible AI Use in Government” as unverified in status until you can see it issued on an official source; a government-use framework, if issued, would address agencies rather than private firms.
  • No Philippine AI risk-tier classification exists. The EU AI Act’s prohibited/high-risk structure is foreign law. Borrowing it as a design pattern is legitimate; calling it a Philippine requirement is a misstatement.
  • No general Philippine rule requires labelling content as AI-generated. The nearest binding hooks are DPA transparency where personal data is involved, the AI Washing prohibition in Section 2(C), and, for synthetic depictions of real people, the NPC’s public notice of 11 August 2026 on real persons’ likenesses in AI-generated images.
  • Advisory No. 2024-04 contains no effectivity clause and no transition period — consistent with its function of explaining how existing obligations already apply.
  • It does not define “AI developer” or “AI deployer” as separate regulated roles. Responsibility runs through the controller/processor distinction, not an EU-style value-chain taxonomy.
  • Advisory No. 2026-01 does not expressly address AI model training. It governs scraping of publicly available personal data. Its requirements plainly reach collection used for training, but do not cite it as an AI-training rule in terms it does not use.

For a market-level view of who sells AI in the Philippines, where the data centers that host it sit, and how those choices meet the Data Privacy Act, see our guide on why your business is already accountable for AI.

Boundaries and Sector Overlays

The DPA carries its own scope limits, and Advisory No. 2024-04 rides on that scope: no personal data, no Advisory — though other law may still apply.

Regulated entities carry overlays. Institutions supervised by the Bangko Sentral ng Pilipinas are subject to BSP issuances on technology and cyber risk in addition to the DPA. BSP Memorandum No. M-2026-034, reported issued on 6 July 2026, has been described as AI-related cyber-risk guidance for BSP-supervised institutions rather than a general AI governance framework; its text could not be retrieved from bsp.gov.ph on September 6, 2026, so treat that description as unconfirmed and verify scope with your BSP compliance function. Either way, a BSP issuance binds BSP-supervised institutions only — it creates no obligation for a retailer, BPO or manufacturer. The same logic applies to Insurance Commission circulars and to sectoral duties in healthcare, education and government.

Hypothetical Scenarios

The following are hypothetical and illustrative only.

A lender deploys an AI credit screen

The model scores applicants and auto-declines the bottom band. Personal data is processed, so Advisory No. 2024-04 applies. Because the decision has legal effect and is made solely by automated processing, IRR Section 48(b) puts the arrangement in difficulty absent consent. Circular No. 2022-04 Section 5(A) requires registration “in all instances,” and Section 26 requires the record to state the lawful basis, retention period, “methods and logic utilized,” and the decisions that would significantly affect the data subject.

A BPO pilots AI quality monitoring of agents

Recordings and transcripts are employee personal data and the tool scores performance, so the higher-risk classification is obvious. The employer needs a lawful basis, a notice disclosing the factors and inputs the system considers and its expected output, a PIA before deployment, and human review before any disciplinary consequence attaches. The fairness duty means testing whether the model scores certain accents or queues more harshly.

A marketing team scrapes public profiles

Advisory No. 2026-01 applies. It states that “the public availability of personal data does not constitute consent by the data subject to its processing for purposes beyond those reasonably contemplated,” requires a defined legitimate purpose and notice, and mandates a PIA covering the activity including work done by third parties. Large-scale social media profiling is expressly listed among the harms controllers must not cause.

Documentation and Records

If the NPC opens an inquiry, these artefacts decide whether your framework was real:

  • Records of processing activities describing each system and identifying who has access — IRR Section 26(c).
  • Privacy Impact Assessments dated before deployment, with the data-flow description, adherence assessment, risk analysis, proposed measures and stakeholder involvement — Advisory No. 2017-03; Circular No. 2023-06 Section 5.
  • Registration records and, for ADM or profiling systems, the notification content required by Circular No. 2022-04 Section 26.
  • Model documentation — purpose, inputs, factors considered, expected output, known limitations, bias testing and results, retraining and scrubbing records.
  • Human-oversight logs — who reviewed what, when, what they overturned, and evidence of reviewer competence and authority; plus records of contested decisions and their resolution.
  • Consent and notice records — the notice version in force at the time, and where consent is the basis, the form or manner of obtaining it.
  • Vendor documentation — processing agreements, subprocessor lists, security assurance, transfer arrangements, diligence findings.
  • Security and access logs — authentication attempts and security incidents, retained longer under Circular No. 2023-06 Section 29.
  • Governance minutes — approvals of higher-risk deployments and the periodic effectiveness review.

Executable Procedure

  1. Read the controlling issuance in full — NPC Advisory No. 2024-04. It is short, and it is what will be quoted back at you.
  2. Confirm your DPO is designated and registered — IRR Section 26(a); Circular No. 2022-04 Section 8.
  3. Build the AI inventory and classify each entry against the Advisory’s threshold: does the automated processing pose significant risk to rights and freedoms?
  4. Run a PIA before deployment per Advisory No. 2017-03, forgoing it only where you can document minimal risk.
  5. Fix the lawful basis under RA 10173 Sections 12 and 13 before processing begins — Advisory Section 2(F).
  6. Rewrite the privacy notice to disclose purpose, the factors and inputs the AI considers, associated risks, expected output, impact on data subjects and dispute mechanisms.
  7. Install human oversight for higher-risk systems, build the contest channel, and register ADM and profiling systems with the Section 26 notification content.
  8. Apply the security floor in Circular No. 2023-06.
  9. Paper the vendors, including transfer arrangements. Model clauses are optional, not an approval route.
  10. Publish the policy and train staff — start from our free AI usage policy template and ChatGPT and AI at work.
  11. Diarise the effectiveness review and minute the result.

Consequences and Realistic Outcomes

NPC enforcement. The Commission investigates complaints and may act on its own initiative, issuing compliance and cease-and-desist orders — the mechanism it described in its 11 August 2026 notice on AI-generated likenesses. The common outcomes are orders to stop or modify processing, corrective directives and monitoring, not immediate criminal prosecution.

Criminal penalties under RA 10173. Sections 25 to 32 attach to defined offences, not to a general failure to govern AI well. Unauthorized processing of personal information carries one to three years’ imprisonment and a fine of ₱500,000 to ₱2,000,000; where sensitive personal information is involved, three to six years and ₱500,000 to ₱4,000,000. Negligent access, improper disposal, processing for unauthorized purposes, concealment of a breach and unauthorized disclosure each carry their own ranges. Section 34 places liability on responsible officers, and Section 35 applies maximum penalties where at least 100 persons are affected.

What will not happen. There is no Philippine AI-specific administrative fine, because there is no Philippine AI statute. Anyone quoting a percentage-of-turnover penalty is describing the EU AI Act, which is foreign law.

Sector consequences run separately. A BSP-supervised institution faces supervisory action under the BSP’s own framework in addition to anything the NPC does — different standards, different timelines, and only for supervised entities.

Practical exposure. For most businesses the realistic near-term risks are commercial: a client audit that finds no PIA, procurement disqualification, or a complaint that surfaces an unregistered profiling system.

Common Mistakes

  • Waiting for the AI law. The DPA already applies; there is nothing to wait for.
  • Importing EU risk tiers as Philippine requirements. Borrow the structure, do not label it local law.
  • Treating a filed bill as a deadline. Label bills as bills in every internal paper.
  • Ceremonial human review. A reviewer without authority to overturn does not meet the Advisory’s standard.
  • Assuming the vendor’s compliance is yours. The controller stays accountable for its processors.
  • Missing registration for profiling systems. Section 5(A) says “in all instances,” with no size threshold.
  • Marketing overstatement. AI Washing is expressly prohibited by Section 2(C).
  • Governing only sanctioned tools. Shadow AI is still processing you are accountable for, and the PIA obligation follows the processing, not the procurement route.

Implementation Checklist

Assign every line to a named person. Unowned and undated items are the ones that fail an inquiry.

# Control Owner Anchor
1 DPO designated and registered with the NPC Management / DPO IRR s.26(a); Circular 2022-04 s.8
2 AI inventory complete, including employee-introduced tools DPO with IT IRR s.26(c)
3 Use cases classified against the significant-risk threshold DPO with business owner Advisory 2024-04 s.2(B)
4 PIA completed and dated before deployment DPO Advisory 2017-03; Circular 2023-06 s.5
5 Lawful basis determined and documented per use case Legal / DPO RA 10173 ss.12-13
6 Notice discloses AI factors, inputs, output, risks, dispute route Legal Advisory 2024-04 s.2(A)
7 Named reviewers with competence and authority to overturn Business owner Advisory 2024-04 s.2(B)
8 Contest channel for automated decisions, with response time Customer / HR lead Advisory 2024-04 ss.2(B), 3(A)
9 Registration and ADM/profiling notification filed DPO Circular 2022-04 ss.5(A), 26
10 Bias identification and monitoring, results recorded Data lead Advisory 2024-04 s.2(C)
11 Accuracy controls and default exclusion of non-contributing personal data Data owner / Engineering Advisory 2024-04 ss.2(D), 2(E)
12 Access control, MFA, logging and secure disposal in place IT security Circular 2023-06 ss.13, 16, 29, 30
13 Vendor agreements, subprocessor list, transfer arrangements Procurement / Legal Advisory 2024-04 s.2(B); Advisory 2024-01
14 Scraping assessed where training or enrichment data is collected Data lead Advisory 2026-01
15 AI usage policy issued, staff trained, incident escalation route defined HR / DPO / IT security Circular 2023-06 s.4(F)
16 Periodic effectiveness review of the framework, minuted Management Advisory 2024-04 s.2(B)

Frequently Asked Questions

Is there an AI law in the Philippines?

No. As of September 6, 2026 there is no comprehensive Philippine AI statute. Bills have been filed but none has become law. See our overview of AI law in the Philippines.

Does Advisory No. 2024-04 apply if we only use AI for drafting?

It applies where personal data is processed. Drafting on non-personal information sits outside its scope, but the moment employee names, customer details, transcripts or CVs enter a prompt, the Advisory and the DPA apply. Our companion guide on AI and data privacy covers those rules.

Are we required to register with the NPC?

Registration is mandatory under Circular No. 2022-04 Section 5 if you employ 250 or more persons, process sensitive personal information of 1,000 or more individuals, or process data likely to pose a risk to rights and freedoms. Separately, Section 5(A) requires registration in all instances where a system involves automated decision-making or profiling. See our data privacy compliance checklist.

Does BSP Memorandum No. M-2026-034 apply to us?

Only if you are a BSP-supervised institution. Reported on 6 July 2026 as AI-related cyber-risk guidance for supervised institutions rather than a general national AI framework, its text could not be retrieved for direct verification on September 6, 2026. Confirm its scope with the BSP before relying on it.

Official Sources

Disclaimer

Important: This article provides general educational information about Philippine law, regulation, cybersecurity, technology, or business compliance. It is not legal advice and does not create an attorney-client relationship. Laws, agency procedures, technical standards, platform rules, and the facts of each situation may change the result. Verify current requirements through the cited official sources and seek qualified professional advice when your rights, deadlines, money, safety, or legal exposure may be affected.

CyberCode updates

Get practical updates on Philippine technology law, data privacy, cybersecurity, and AI.

Email activity tracking

Unsubscribe any time. See our privacy policy below.