CyberCode.ph · Philippines

AI and Data Privacy Philippines: Personal Data Rules for AI Systems and Generative AI

Last updated October 3, 2026 · Practical privacy, cybersecurity and technology-law guidance
Padlock and keyboard representing AI data privacy and personal data protection in the Philippines

Last materially reviewed: September 6, 2026

Direct Answer

The Philippines has no comprehensive AI statute. What binds an AI system here is the Data Privacy Act of 2012 (RA 10173), its Implementing Rules and Regulations, and National Privacy Commission issuances — applied unchanged to AI. NPC Advisory No. 2024-04 (19 December 2024) states the position directly: those rules are “principles-based and technology-neutral” and apply whenever personal data is processed in developing or deploying an AI system, including its training and testing.

For related governance, workplace, security, deepfake and intellectual-property questions, continue to Cybercode’s Philippine AI and emerging-technology hub.

Key Takeaways

  • Using AI changes nothing about your Data Privacy Act obligations. It only changes how visibly you fail them.
  • You must determine the lawful basis under RA 10173 Sections 12 and 13 before processing begins — Advisory No. 2024-04, Section 2(F).
  • Publicly available personal data “do not lose the protection afforded by the DPA” simply because it is public.
  • IRR Section 48(b): no decision with legal effects may be made solely on automated processing without the data subject’s consent.
  • IRR Section 48(a) requires notification to the NPC where automated processing is the sole basis for a significant decision, including the “methods and logic utilized.”
  • Rights must be exercisable “before, during, and after” development or deployment, and being in a training dataset “does not automatically make the exercise of data subject rights unreasonable” — Section 3.

AI privacy decision map

Question Practical answer
Does the Data Privacy Act apply to AI training and testing? Yes when personal data is processed. NPC Advisory No. 2024-04 applies the technology-neutral DPA framework to AI development and deployment, including training and testing.
Does public personal data lose Data Privacy Act protection? No. Public availability does not by itself remove DPA protection or the need for a lawful purpose and appropriate safeguards.
When should an AI project undergo a privacy impact assessment? Run a PIA before deployment where the processing creates meaningful privacy risk; NPC Advisory No. 2026-01 specifically requires a PIA for covered data-scraping activity.
What about solely automated significant decisions? The DPA IRR imposes additional rules for automated processing used as the sole basis of decisions with significant legal effects, including consent and NPC-notification requirements described in Section 48.
What should a business document? Purpose, lawful basis, data sources, training/testing use, retention, security, vendor roles, human oversight, rights handling and the PIA or risk assessment.

Jump to a Section

What Actually Binds AI Processing

Status as of September 6, 2026. Every row below is an issued instrument, not a proposal.

Instrument Issuer Date Who it binds What it requires for AI
RA 10173, Data Privacy Act of 2012 Congress 2012 PICs and PIPs processing personal data Ss.11 principles; 12-13 lawful basis; 16 data subject rights; 20 security; 21 accountability
IRR of RA 10173 (as amended) NPC As amended PICs and PIPs S.26(a) DPO; s.26(c) records of processing; s.34 rights; s.48 automated processing notification and the solely-automated rule; s.50 accountability for transfers
NPC Advisory No. 2024-04 NPC 19 Dec 2024 PICs and PIPs where AI processes personal data, including training and testing Transparency, accountability, fairness, accuracy, minimisation, lawful basis, exercisable data subject rights
NPC Circular No. 2022-04 NPC 5 Dec 2022 Covered PICs and PIPs; all ADM or profiling systems Registration; ADM/profiling notification content; DPO designation
NPC Circular No. 2023-06 NPC 1 Dec 2023 PICs and PIPs Security measures, PIA, privacy-by-design, access control, logging, retention, disposal
NPC Advisory No. 2017-03 NPC 31 Jul 2017 PICs and PIPs PIA before adoption, unless risks are minimal
NPC Advisory No. 2026-01 NPC 13 Apr 2026 Those scraping publicly available personal data, and PICs hosting it Purpose, lawful basis, notice, minimisation, mandatory PIA, sensitive-data limits, source disclosure
NPC Advisory No. 2024-01 NPC 30 May 2024 Parties to cross-border transfers Model contractual clauses — voluntary, not mandatory
NPC Notice to the Public NPC 11 Aug 2026 Creators and platforms circulating AI images of real people Position that unauthorised synthetic likenesses are false personal data engaging DPA liability

Governing Law and Authorities

RA 10173 and its IRR are the source of obligation; the National Privacy Commission is the regulator. Advisory No. 2024-04 is the interpretive bridge. Its Section 1 sets scope: “This Advisory applies when the processing of personal data is involved in the development or deployment of AI systems, including its training and testing.”

Section 2 then states the general rule that answers most questions people ask about AI and privacy: the obligations of controllers and processors “including adhering to the general privacy principles, upholding the rights of the data subjects, determining the appropriate lawful basis for processing, and implementing security measures, apply insofar as AI systems involve the processing of personal data.” There is no AI carve-out, and no AI-specific consent regime.

The Advisory defines an AI system broadly, covering systems that “learn from data, perceive environments, process and adapt to new inputs, manipulate or generate output.” A generative assistant, a scoring model and a recommendation engine all fall inside it.

The Obligations, Element by Element

When AI processing becomes personal data processing

The trigger is the data, not the technology. Prompts containing employee or customer details, uploaded CVs, call transcripts, images of identifiable people, inferred attributes and outputs about a named individual are all personal data. So is training data: the Advisory reaches development as well as deployment, so a dataset assembled for fine-tuning is squarely in scope.

Lawful basis

Section 2(F) requires controllers to “determine the most appropriate lawful basis under Sections 12 and 13 of the DPA prior to the processing.” So the basis must be selected before the model touches the data, not reconstructed after a complaint. Consent is one option among several under Section 12; Section 13 governs sensitive personal information and is materially narrower.

The Advisory closes the most common loophole directly: “Publicly available personal data do not lose the protection afforded by the DPA… simply because it has been made public or is publicly accessible.” A controller processing public data must still identify a lawful basis and adhere to the general privacy principles.

Transparency and privacy notices

Section 2(A) requires controllers to inform data subjects of “the nature, purpose, and extent of the processing.” For AI, the Advisory specifies what must be disclosed: the purpose for processing, the “factors and inputs considered by such AI systems,” associated risks, expected AI output, impact on data subjects, and available dispute mechanisms.

It also sets a comprehension standard: the information must be “easy to access, concrete and definitive, understood by members of their target audience, and presented in a simple manner using clear and plain language.” A notice saying only “we may use artificial intelligence” does not meet it. IRR Section 34(a) separately gives the data subject a right to be informed of processing “including the existence of automated decision-making and profiling.”

Automated decision-making and profiling

This is where Philippine law is most concrete, and where most AI deployments carry unrecognised exposure. IRR Section 48(b) is a hard rule: “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.” Section 48(a) requires notification to the Commission where automated processing becomes the sole basis for a decision that would significantly affect the data subject, covering the purpose, categories of data and data subject, consent forms or manner of obtaining consent, recipients, retention period, the “methods and logic utilized for automated processing,” the decisions to be made, and the DPO’s contact details.

NPC Circular No. 2022-04 operationalises this. Section 5(A): “A Data Processing System processing personal or sensitive personal information involving automated decision-making or profiling shall, in all instances, be registered.” Section 26 requires the record to state the lawful basis, retention period, methods and logic, and the possible decisions affecting the data subject. There is no size threshold — it applies to a ten-person company as much as a bank.

RA 10173 Section 16(c) reinforces the access right, entitling the data subject to information on automated processes “where the data will or likely to be made as the sole basis for any decision significantly affecting” them. Advisory No. 2024-04 adds operational safeguards, covered in our AI governance framework guide: meaningful human intervention by competent, authorised persons, and a route to question and contest.

Data subject rights against AI systems

Section 3 is unusually practical. Controllers “shall implement effective mechanisms or alternative measures to give effect particularly to the right to object, the right to rectification, and the right to erasure or blocking,” exercisable “before, during, and after the development or deployment of AI systems, including its training and testing.”

It also anticipates the standard industry defence and rejects it. Incorporation into datasets “does not automatically make the exercise of data subject rights unreasonable,” and controller “inaction towards providing mechanisms that allow for the meaningful exercise of data subject rights negates any claim that fulfilling such requests is unreasonable.” Where a request genuinely cannot be fulfilled, the controller must say so with reasons and still achieve the intended effect “as far as possible.”

Minimisation, accuracy and fairness

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

Training data and scraping

NPC Advisory No. 2026-01 (13 April 2026) governs scraping of publicly available personal data. It does not use the words “model training,” so do not cite it as an AI-training rule, but its requirements plainly reach data collected for that purpose. It requires a specific and legitimate purpose, a lawful basis, notice “before the processing takes place, or at the next practical opportunity,” minimisation, privacy-by-design, and a mandatory PIA “covering such activities, including those performed for and on its behalf by third parties.” Scraping sensitive personal information is prohibited unless the controller shows a Section 13 basis, strict necessity and proportionality, and enhanced security. Where scraped data is obtained from someone else, Section 7 requires policies verifying the source complied with the DPA, and disclosure of the source in the privacy notice.

Vendors, processors and cross-border transfer

A third-party AI provider usually acts as a personal information processor, and the controller stays accountable: Advisory No. 2024-04 says controllers “shall remain accountable for the actions of their PIPs,” and IRR Section 50 states the same for transfers. Where processing happens abroad, NPC Advisory No. 2024-01 (30 May 2024) publishes model contractual clauses drawn from ASEAN, EU, UK and other frameworks — but use is voluntary: “The NPC does not require contractual parties to adopt the use of MCCs,” and it “will not accept any requests for review of agreements or contracts to determine conformity.” See our AI vendor due diligence checklist. The same accountability applies when AI arrives inside ordinary office equipment, such as cloud-based AI features in Canon printers.

Security, PIA and registration

NPC Circular No. 2023-06 (1 December 2023) sets the security baseline: need-to-know access control with authentication (Section 13), multifactor or similar authentication for online access to sensitive data (Section 16), encryption on removable media (Section 25), documented retention periods (Section 9), longer retention of authentication and incident logs (Section 29) and secure disposal (Section 30). NPC Advisory No. 2017-03 (31 July 2017) is the PIA instrument: a PIA should be undertaken for every processing system involving personal data, new and existing, before adoption for new systems, and may be forgone only where the controller determines the processing “involves minimal risks.”

Synthetic images of real people

In a public notice dated 11 August 2026, the NPC addressed AI-generated images and videos depicting real, identifiable people. It treats face and likeness as personal information, characterises synthetic depictions of things a person never did as false personal data engaging the RA 10173 Section 16(e) right to demand blocking, removal or destruction, and points to Section 25 for liability. Satire, parody, commentary and journalism are narrow exceptions only where the synthetic character is disclosed. Platforms are asked to provide accessible reporting and prompt removal, with priority for complaints involving minors.

What Is Not Settled or Not Yet Issued

As of September 6, 2026:

  • There is no Philippine AI statute and no AI-specific data protection law. Bills have been filed; a filed bill is not law and imposes no obligation.
  • No issued 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, still working to develop “the first national AI governance framework.”
  • There is no Philippine “right to explanation” in the EU sense. What exists is narrower and more specific: the access right in RA 10173 Section 16(c) to information on automated processes that are the sole basis of a significant decision, the notification duty in IRR Section 48(a), and the registration content in Circular No. 2022-04 Section 26. Do not promise clients a GDPR-style explainability right under Philippine law.
  • Advisory No. 2024-04 does not define “AI developer” or “AI deployer” as distinct regulated roles, and does not define “training data.” Responsibility runs through the controller/processor distinction.
  • Advisory No. 2026-01 does not expressly mention AI or model training, and Advisory No. 2024-04 has no effectivity clause or transition period, because it explains obligations already in force rather than creating new ones.
  • The NPC does not pre-approve contracts. Advisory No. 2024-01 says so expressly. There is no route to obtain NPC sign-off on an AI vendor agreement or a transfer arrangement.
  • There is no general statutory rule on AI-content labelling. The 11 August 2026 notice addresses likenesses of real people under the DPA; it is not a general synthetic-media labelling law.

Exceptions, Boundaries and Sector Overlays

The DPA carries its own scope limits and exclusions, and everything above rides on that scope. A genuinely anonymised dataset that cannot reasonably identify anyone is not personal data — but re-identification risk is easy to underestimate, and the burden of that assessment sits with you.

Sector overlays sit on top of the DPA rather than replacing it. Institutions supervised by the Bangko Sentral ng Pilipinas remain subject to BSP technology and cyber-risk issuances alongside the NPC’s rules. 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 framework; its text could not be retrieved from bsp.gov.ph on September 6, 2026, so treat that description as unconfirmed and verify with the BSP. It binds supervised institutions only. Insurance Commission circulars bind insurers; government agencies carry additional duties, including under RA 10173 Section 22.

Hypothetical Scenarios

The following are hypothetical and illustrative only.

An HR team uploads CVs to a generative assistant to shortlist candidates

CVs are personal data and may contain sensitive personal information, so a Section 12 basis — or Section 13 where sensitive data is involved — must be fixed before upload, and the applicant notice must disclose the factors and inputs the system considers and its expected output. If the recruiter simply accepts the ranking, the decision is effectively solely automated and IRR Section 48(b) bites. Because the system profiles, Circular No. 2022-04 Section 5(A) requires registration regardless of company size.

A retailer fine-tunes a model on five years of customer support transcripts

This is development-stage processing, squarely within Advisory No. 2024-04. The retailer must fix a lawful basis, exclude by default personal data unlikely to improve the model (Section 2(E)), run a PIA before training, and let customers object, rectify or erase — including after training, per Section 3(B). “It is already in the model weights” is not, by itself, an answer the Advisory accepts. Had the training set been bought or scraped, Advisory No. 2026-01 Section 7 would also require the retailer to verify that its source complied with the DPA and to disclose that origin in its privacy notice.

A marketing agency generates an ad using a local celebrity’s face

The NPC’s 11 August 2026 notice treats face and likeness as personal information and unauthorised synthetic depictions as false personal data. Without a lawful basis the agency faces DPA exposure and the individual may demand blocking, removal or destruction. Our guide on who owns AI-generated content covers the separate IP questions.

Documentation and Records

What a regulator would actually ask to see:

  • Processing inventory under IRR Section 26(c) — each AI system, its data flow, and who has access.
  • Lawful basis record per processing activity, with the Section 12 or 13 ground identified and dated before processing began.
  • PIA records — data-flow description, adherence assessment, risk analysis, proposed measures, stakeholder involvement, and the review history.
  • Privacy notice versions as published at each point in time, showing the Section 2(A) disclosures.
  • Consent records where consent is the basis — the form used or the manner of obtaining consent, as Circular No. 2022-04 Section 26 requires for ADM systems.
  • Registration and ADM notification filings and their currency.
  • Model and training-data documentation — sources, whether any data was scraped, minimisation decisions, bias testing, retraining.
  • Rights request log — objections, rectifications, erasure and blocking requests, what was done, and where refused, the reasons and the alternative measures applied; plus human-oversight and contest records for automated decisions.
  • Processor and transfer documentation — agreements, subprocessors, processing locations, security assurance — plus access, authentication and incident logs (Circular No. 2023-06 Section 29), the retention schedule and disposal records.

Executable Procedure

  1. Read NPC Advisory No. 2024-04 in full. It is the instrument you will be measured against.
  2. Map the personal data entering every AI system — prompts, uploads, integrations, training sets and outputs about identifiable people — and fix the lawful basis under RA 10173 Sections 12 and 13 before processing starts.
  3. Run a PIA per NPC Advisory No. 2017-03 before deployment, and again on material change.
  4. Rewrite the privacy notice to carry the Section 2(A) disclosures: purpose, factors and inputs, risks, expected output, impact, dispute mechanisms.
  5. Test whether any decision is solely automated. If it has legal effects, IRR Section 48(b) requires the data subject’s consent; otherwise insert genuine human intervention.
  6. File the ADM notification and register the system under NPC Circular No. 2022-04, including methods and logic.
  7. Build rights mechanisms that work before, during and after training — objection, rectification, erasure and blocking — and log refusals with reasons.
  8. Assess training and enrichment data against NPC Advisory No. 2026-01 if any of it was scraped or bought.
  9. Apply the security baseline in NPC Circular No. 2023-06, and paper processors and transfers. Model clauses under Advisory No. 2024-01 are optional and are not NPC approval.
  10. Train staff on prompt hygiene — see ChatGPT and AI at work and the AI usage policy template.

Consequences and Realistic Outcomes

NPC enforcement. The Commission investigates complaints and may act on its own initiative, issuing compliance orders and cease-and-desist orders — the enforcement route it set out in its 11 August 2026 notice. In practice the realistic outcome of a proceeding is an order to stop or modify processing, corrective directives and monitoring.

Criminal penalties under RA 10173. Sections 25 to 32 create specific offences. Unauthorized processing of personal information carries one to three years and ₱500,000 to ₱2,000,000; for sensitive personal information, three to six years and ₱500,000 to ₱4,000,000. Processing for unauthorized purposes, negligent access, improper disposal, concealment of a breach and unauthorized disclosure each carry their own ranges. Section 34 places liability on responsible officers; Section 35 applies maximum penalties where at least 100 persons are affected. These are criminal provisions prosecuted through the courts, not fines the NPC levies administratively.

What will not happen. There is no AI-specific Philippine penalty, because there is no AI statute. Percentage-of-turnover figures come from the EU AI Act and GDPR — foreign law that does not apply to Philippine operations of its own force.

Sector consequences are separate. A BSP-supervised institution can face supervisory action under BSP rules in addition to anything the NPC does. Those consequences apply only to supervised entities.

Common Mistakes

  • Treating public data as free data. The Advisory says public availability does not remove DPA protection, and Advisory No. 2026-01 says it is not consent.
  • Choosing the lawful basis after deployment. Section 2(F) requires it before processing — and consent is only one Section 12 ground, often the weakest for employment.
  • A notice that says only “we use AI.” Section 2(A) requires factors, inputs, risks, expected output, impact and dispute route, in plain language.
  • Assuming a human in the loop cures a solely-automated decision. If the human always accepts the output, the decision is still effectively automated.
  • Missing registration. ADM and profiling systems must be registered “in all instances” — no size threshold.
  • Claiming erasure is impossible once data is in the model. The Advisory rejects that as an automatic answer and requires alternative measures as far as possible.
  • Assuming the vendor’s certification is your compliance. The controller remains accountable for its processors.

AI Processing Compliance Checklist

# Check Anchor
1 Every AI system with personal data identified and in the processing inventory IRR s.26(c)
2 Lawful basis fixed and recorded before processing began, with sensitive data separately justified RA 10173 ss.12-13; Advisory 2024-04 s.2(F)
3 Privacy notice discloses factors, inputs, risks, expected output, impact, dispute route Advisory 2024-04 s.2(A)
4 Notice states the existence of automated decision-making and profiling IRR s.34(a)
5 No decision with legal effects made solely by automation without consent IRR s.48(b)
6 NPC notified of automated processing that is the sole basis of a significant decision IRR s.48(a)
7 ADM/profiling system registered, with methods and logic stated Circular 2022-04 ss.5(A), 26
8 Rights mechanisms work before, during and after training Advisory 2024-04 s.3(B)
9 Objection, rectification, erasure and blocking routes operational and logged, with refusals answered by reasons plus alternative measures Advisory 2024-04 s.3(A)
10 PIA completed before deployment and reviewed on material change Advisory 2017-03; Circular 2023-06 s.5
11 Non-contributing personal data excluded by default from training Advisory 2024-04 s.2(E)
12 Bias identification and monitoring in place Advisory 2024-04 s.2(C)
13 Scraped or purchased data assessed, source verified and disclosed Advisory 2026-01 ss.3, 7
14 Processor agreements and cross-border arrangements documented IRR s.50; Advisory 2024-01
15 Security baseline implemented: access control, authentication, logs, disposal Circular 2023-06 ss.13, 16, 29, 30
16 DPO designated and registered IRR s.26(a); Circular 2022-04 s.8

Frequently Asked Questions

Is there a Philippine law specifically about AI and privacy?

No separate statute. The Data Privacy Act and its IRR apply, and NPC Advisory No. 2024-04 explains how they apply to AI systems processing personal data.

Can we train a model on data we scraped from public profiles?

Not without meeting NPC Advisory No. 2026-01. You need a specific legitimate purpose, a lawful basis, notice, minimisation and a PIA. Public availability is not consent, and scraping sensitive personal information is prohibited unless you can show a Section 13 basis, strict necessity and enhanced security.

Does a data subject have a right to have their data removed from a trained model?

Advisory No. 2024-04 requires effective mechanisms for erasure or blocking and says inclusion in a dataset does not automatically make a request unreasonable. Where fulfilment is genuinely infeasible you must say so with reasons and still achieve the intended effect as far as possible. No Philippine authority holds that model retraining is always required.

Do we have to register with the NPC if we use an AI tool?

If the system involves automated decision-making or profiling, yes — Circular No. 2022-04 Section 5(A) requires registration in all instances. Otherwise the Section 5 thresholds apply: 250 or more employees, sensitive personal information of 1,000 or more individuals, or processing likely to pose a risk to rights and freedoms. See our data privacy compliance checklist.

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.