AI Agents in PH Government: The DICT–Google Cloud Deal's Legal Questions

1What happened, and where things stand in November 2026

On 21 June 2026, Google Cloud announced an expanded, multi-year collaboration with the Department of Information and Communications Technology (DICT) covering artificial intelligence (AI) for public servants, AI "agents" for citizen services, a shared cybersecurity platform and subsea-cable connectivity. Philippine outlets reported it the following days as a multi-year deal to expand government use of AI and strengthen cybersecurity (Inquirer, 25 June 2026), and foreign trade press, including ComputerWeekly, described it as the Philippine government tapping Google Cloud to deploy AI agents.

Since June, three developments have shaped how the arrangement will play out:

• 13 July 2026 — Executive Order No. 119. The President signed a new government data classification and residency framework, replacing a 1964-era framework and setting rules on where classified government data may be stored, including data held by private contractors.
• 6 August 2026 — 2027 budget request. GMA News reported that the DICT would seek about P18 billion for 2027, including P2.7 billion for e-government programs, with funding linked to cloud hosting, cybersecurity and the planned rollout of "eGovAI".
• 21 September 2026 — eGovAI launch. The DICT unveiled eGovAI in the eGovPH Super App, including an AI assistant called "Kuya AI" for information, navigation of services, identity verification, document processing and translation.

As of the research date (6 November 2026), we could not find published reporting that names the AI vendor or model behind eGovAI, or that confirms which citizen-facing agents built under the Google Cloud program are already live. The public sources also do not disclose the value or contract terms of the Google Cloud arrangement. Those gaps matter for the legal analysis below, and any business relying on this article should confirm the current position directly with the DICT.

• 50,000+
  public servants given initial access to Gemini Enterprise (DICT–Google Cloud, June 2026)
• 200,000+
  public servants targeted within 18 months
• 56 → 90
  agencies onboarded to the shared cyber defence platform, with 90 targeted by end-June 2026
• 120 days
  deadline under EO 119 for implementing guidelines, as reported by the Inquirer

2What the DICT–Google Cloud arrangement actually covers

According to the Google Cloud announcement and GovInsider's report, the collaboration has four strands.

AI tools for public servants

Gemini Enterprise and Google Workspace are offered to public servants through the eMarketplace of the Procurement Service of the Department of Budget and Management (PS-DBM), which launched in the first quarter of 2026. More than 50,000 public servants receive access initially, with a target of over 200,000 within 18 months. Examples include a building official searching pending permit applications. ComputerWeekly reported that the DICT will track usage frequency, cost savings and productivity to measure return on investment.

AI agents for citizens

The DICT plans to use Google's agent platform to build citizen-facing agents integrated into e-government platforms, answering queries in local languages by voice or text. Examples include business registration, health centre schedules and disaster relief guidelines. These are described as plans, not completed deployments.

Cybersecurity

"Google Cloud Cybershield" was deployed at the DICT's National Security Operations Center, combining threat intelligence, Gemini-assisted security operations and Mandiant expertise. Security teams from 56 agencies had been onboarded, with 90 targeted by the end of June 2026, and the same architecture supports digital operations for the ASEAN Summits hosted by the Philippines from April to November 2026.

A fourth strand covers subsea cable capacity linked to the DICT's Luzon Bypass Infrastructure and Free Wi-Fi for All program.

“Transformation is not defined by the adoption of sophisticated software tools alone”

— DICT Secretary Henry Rhoel Aguda, Google Cloud press release, 21 June 2026

3Data privacy: the agency stays accountable

Government agencies are personal information controllers under the Data Privacy Act of 2012 (Republic Act No. 10173). Bringing in a cloud or AI provider does not move that responsibility. The vendor will usually be a personal information processor acting on the agency's instructions, and the agency must bind it by contract to the security and confidentiality standards the law requires.

The National Privacy Commission (NPC) spelled out how these rules apply to AI in NPC Advisory No. 2024-04 (19 December 2024), which covers personal data processing in developing or deploying AI systems, including training and testing. Several of its points bear directly on citizen-facing agents:

• Transparency. Data subjects should be told the nature, purpose and extent of processing, and the advisory expects controllers to explain the system's inputs, risks, expected outputs and the dispute mechanisms available.
• Accountability for processors. Controllers remain accountable for AI processing outcomes, including work done by processors, and must keep documentation showing compliance.
• Governance tools. The advisory points to privacy impact assessments, privacy-by-design and by-default, continuous monitoring and a dedicated AI ethics board.
• Fairness and accuracy. Processing must not be manipulative or unduly oppressive; controllers should identify and monitor bias, and the advisory warns against "AI washing," or overstating AI involvement.
• Data minimisation. Personal data unlikely to improve the AI system should be excluded by default.
• Rights. Rights to object, rectification and erasure or blocking remain available before, during and after deployment, and including data in a training set does not by itself make a request unreasonable.

In practice, a privacy impact assessment should be done per use case, not once per product: an internal tool that summarises policy papers carries very different risk from an agent that reads an applicant's records. Agencies and their vendors should also settle whether citizen prompts, uploaded documents and conversation logs may be retained by the provider or used to improve its models; the public announcements do not address this.

Agencies already handling personal data at scale will also have existing obligations on breach management. Our earlier guide on Data Privacy Act breach notification duties for DPOs applies equally when the system involved is an AI agent.

4Automated decisions and who answers for them

The Data Privacy Act gives data subjects a right to be informed of automated processes where data will be used as the sole basis for a decision significantly affecting them. The NPC's AI advisory builds on this: where automated decisions pose significant risk, controllers should provide "meaningful human intervention" by persons with the competence and authority to change the outcome, and data subjects should be able to question and contest the decision.

That distinction is central. An agent giving a health centre's opening hours provides information; an agent that screens an application for disaster relief, flags a business registration as incomplete, or ranks permit requests is closer to making or shaping an administrative decision. In that second category, several legal principles meet:

• Administrative due process. Decisions affecting rights and benefits must rest on evidence, give the affected person a chance to be heard, and be explainable to a reviewing authority. A response generated by a model that cannot show its basis is difficult to defend on review.
• Accountability of officials. Liability under public-accountability laws attaches to officials and employees, not to software. Agencies should document which official owns each AI-assisted process and signs off on outputs that affect individuals.
• Accuracy and language. Answers in local languages are a real benefit, but translation errors in eligibility rules or deadlines can cause harm. Agencies should log agent responses and keep the authoritative rule text human-maintained.

If later phases let agents act on applications, agencies will need written human-review protocols, appeal routes and audit trails before go-live. As of 6 November 2026, we have not seen a published DICT policy on human oversight specific to these agents.

5Cloud, data residency and EO 119

The June announcements say nothing about where citizen data processed by the AI agents will be stored. Executive Order No. 119 now provides the framework for that question for government data. Based on reporting by PTV News, the Inquirer and Baker McKenzie:

The order was signed on 13 July 2026 and, according to Baker McKenzie, took effect on 15 July 2026 upon publication. It covers government data held by national agencies, government-owned or -controlled corporations and state universities and colleges, and reaches government data processed or stored by private entities on behalf of an agency, including in public-private partnerships, utilities and critical infrastructure. Purely private data is outside its scope. A Joint Oversight Committee co-chaired by the DICT and the National Security Council is to issue implementing guidelines; the Inquirer reported a 120-day deadline and a three-year phased compliance timetable.

If counted from the reported 15 July effectivity date, the 120-day window ends in mid-November 2026. As of 6 November 2026, we had not confirmed whether the guidelines have been issued. Their content will decide many practical questions for AI vendors, such as how agencies classify the data an agent touches, whether logs and prompts inherit the classification of the records they reference, and how offshore backups and disaster-recovery copies are treated.

EO 119 deals with government data classification. It does not replace the Data Privacy Act, and personal data in a citizen's request remains protected by the DPA wherever it sits. Vendors therefore need to comply with both: EO 119's residency rules for the government data, and the DPA and NPC rules for the personal data within it.

6Procurement of AI under RA 12009

The New Government Procurement Act, Republic Act No. 12009, was signed on 20 July 2024 and took effect on 13 August 2024; its implementing rules were approved in February 2025. The Government Procurement Policy Board's NGPA primer lists the use of emerging technologies and innovative solutions among the law's principles and names new modalities including competitive dialogue, direct procurement for science, technology and innovation, framework agreements, unsolicited offers with bid matching, and life-cycle costing.

The Google Cloud tools reach public servants through the PS-DBM eMarketplace, which GovInsider reports is itself built on Google Cloud infrastructure. The public reports do not explain how the arrangement was procured or priced. That is not a finding of irregularity; it is simply information the sources do not provide, and readers interested in the specifics should check PhilGEPS postings and agency disclosures.

For AI procurement generally, the law's tools suit the problem better than the older regime did:

• Competitive dialogue allows an agency to refine requirements with bidders where the best technical solution is not known in advance, which is often the case with agent design.
• Life-cycle costing lets agencies weigh usage-based AI pricing, storage and exit costs over the contract, not just the first-year licence fee.
• Framework agreements can support repeated, smaller orders, such as adding agents for new services over time.

Our article on public procurement transparency reforms covers the wider compliance framework.

Contract terms worth negotiating

Whatever the modality, AI contracts with government should address: data ownership and use limits (including no use of agency or citizen data to train general models without authority); location of storage, backups and logs consistent with EO 119; breach notification timelines aligned with NPC rules; audit and access rights for the agency and the NPC; model change notices, since providers update models frequently; human-review obligations for consequential outputs; and exit, data return and deletion.

7Cybersecurity: a shared platform, shared risks

Centralising security monitoring for dozens of agencies on one platform can raise baseline protection, but it also concentrates risk: a misconfiguration or compromised credential can affect many agencies at once.

AI agents introduce new attack surfaces. Prompt injection, where hidden instructions in a document or web page redirect an agent, and over-broad permissions that let an agent read or change more records than its task requires, are recognised risks for agentic systems. Agencies should give each agent the narrowest data access it needs, log its actions, and test it against misuse before exposing it to the public. Where an incident involves personal data, NPC breach-notification rules apply, and unauthorised access may also be prosecuted under the Cybercrime Prevention Act.

The DICT's 2027 budget request, as reported by GMA News, ties e-government funding to cloud hosting, cybersecurity and system integration. Whether and how much Congress approves remains part of the 2027 budget process and should be confirmed against the enacted General Appropriations Act.

8What private tech vendors should do now

The DICT's direction is clear: more AI inside agencies and in citizen services, backed by a larger e-government budget request. That creates openings for local integrators, developers and AI specialists, including as subcontractors to large cloud providers, and legal readiness will matter as much as technical capability.

• Map which government data your solution will touch and how it is likely to be classified under EO 119; plan hosting, backups and logs accordingly, and watch for the implementing guidelines.
• Prepare a privacy impact assessment template for AI use cases that follows NPC Advisory No. 2024-04, including bias testing and human-review design.
• Document your model supply chain: which foundation model, where it runs, and whether prompts or outputs are retained or used for training.
• Review your eligibility and documentation under RA 12009 and its IRR, including PhilGEPS registration and the modality the agency is likely to use.
• Build contract positions on data use, audit rights, breach notification, model updates and exit before you are in negotiations.
• Keep corporate, licensing and foreign-ownership records in order; agencies and prime contractors will ask.

For broader context, see our pieces on how AI is changing app development and automation from a legal view and the 2026 update on Philippine cybersecurity and digital governance policy.

• The June 2026 DICT–Google Cloud arrangement covers AI tools for public servants, planned citizen-facing agents, shared cyber defence and connectivity; its value and terms were not disclosed in the public sources.
• The DICT launched eGovAI and "Kuya AI" in the eGovPH app on 21 September 2026; the underlying vendor was not named in the reporting we reviewed.
• Agencies remain accountable under the Data Privacy Act, and the NPC's 2024 AI advisory expects transparency, impact assessments and meaningful human review for high-risk automated decisions.
• EO 119 (13 July 2026) sets residency rules for government data, including data held by private contractors; implementing guidelines are reportedly due around mid-November 2026.
• RA 12009 gives agencies newer procurement tools suited to AI, and vendors should prepare for privacy, residency and contract scrutiny together.

If your company is preparing to bid for, subcontract on, or supply AI or cloud services to a government agency, Abanto Law Firm can review your contracts, data-handling structure and procurement documentation. Learn more about our corporate law services or contact our team to discuss your situation.

9Frequently asked questions