On 30 July 2025, Pakistan's federal cabinet approved the country's first National Artificial Intelligence Policy. Seven months later, in February 2026, Islamabad hosted the signing of the Islamabad AI Declaration — a public statement of intent to build what officials described as sovereign AI capability. For a country that usually imports its technology law a decade after everyone else has finished writing it, this was unusually early.
And yet the national conversation that followed missed the point almost entirely.
Ask a policymaker, a bar-council committee, or a university law faculty what artificial intelligence means for Pakistani law, and nine times out of ten the answer is a debate about copyright: if a machine writes a poem or drafts a contract, who owns it — the user, the developer, or nobody? It is a real question. Pakistan's Copyright Ordinance of 1962 recognises only human authorship and is entirely silent on machines. But it is a small question wearing the costume of a large one.
This article argues that Pakistan is asking the wrong question, and that the right one is constitutional. When a nation's courts, hospitals, tax authorities, universities, and legislatures all begin to run on artificial intelligence that no Pakistani institution built, owns, or can independently audit, the issue stops being about who owns an output. It becomes about who governs the intelligence itself. What follows is a working framework for that harder question — a set of constitutional stress tests, a comparative reading of what the world's AI powers actually did, and an original governance model we believe Pakistan is uniquely placed to author rather than import.
Where Pakistan actually stands in 2026
Begin with the facts, because the facts are more interesting than the slogans.
The National AI Policy 2025 is not a statute. It is a policy instrument approved by cabinet, and it sets up institutions rather than rules: an AI Directorate under the Ministry of Information Technology and Telecommunication, a National AI Fund to seed research and commercialisation, a National AI Council to steer the plan, geographically distributed Centres of Excellence, and — importantly for lawyers — regulatory sandboxes for supervised testing. It is, in effect, an industrial-strategy document with a light-touch governance spine. It does not amend the Constitution, the Copyright Ordinance, the Qanun-e-Shahadat Order, or the Code of Civil Procedure. It does not tell a judge how to treat an AI-drafted submission, or a citizen how to challenge an automated rejection.
Around that policy sits a legal order that predates the technology by generations. Copyright rests on a 1962 ordinance built for human authors. The Personal Data Protection Bill — the single instrument most relevant to how AI systems ingest citizens' information — has been approved by cabinet in successive drafts but, as of mid-2026, has still not been passed by both houses of Parliament. It is not law. Meanwhile the constitutional guarantees that AI will most directly stress are exactly the ones the courts have spent seventy years building: Article 4 (the right of every citizen to be dealt with in accordance with law), Article 9 (liberty), Article 10A (fair trial and due process), Article 14 (the inviolable dignity of man and privacy of the home), and Article 25 (equality before law).
Those articles are not decorative. In Muhammad Rehmatullah v. The State (2023), the Lahore High Court held that extracting data from a smartphone without a judicial warrant violated fundamental rights, treating digital privacy as equivalent to the inviolability of the home under Article 14. A quarter-century earlier, in Benazir Bhutto v. President of Pakistan (1998), the Supreme Court struck down unauthorised phone tapping. Pakistani courts already understand that new technologies do not create rights-free zones. What they have not yet faced is a technology that does not merely surveil citizens but reasons about them.
So the real map of 2026 looks like this: an ambitious industrial policy, a data-protection statute stuck in the legislative pipeline, a copyright law that cannot see machines, and a constitutional tradition strong enough to matter but not yet tested against automated decision-making. That gap — between institutional ambition and constitutional readiness — is where the governance work has to happen.
Why copyright is the wrong question
The copyright debate is seductive because it is tractable. It has familiar parties (author, publisher, infringer), a familiar remedy (damages, injunction), and a familiar forum (the IP tribunal). Lawyers reach for it the way a person searching for lost keys searches under the streetlight — not because the keys are there, but because the light is good.
Consider what the copyright frame cannot reach. Suppose an AI system is trained on every reported judgment of the Supreme Court and the High Courts of Pakistan. It copies nothing verbatim; it learns statistical patterns and produces original prose. Under a 1962 ordinance that protects the expression of works and requires a human author, it is far from clear that anything has been "copied" at all. And if nothing has been copied, then the entire corpus of Pakistani jurisprudence — public records, yes, but the accumulated reasoning of the Republic — has quietly become raw material for a privately owned commercial intelligence product, with no Pakistani institution holding any governance interest in the result.
The copyright question asks: who owns this output? The constitutional question asks something the twentieth century never had to: who owns, governs, preserves, and can switch off the intelligence infrastructure on which the future State will depend? One is a dispute about a poem. The other is a question about sovereignty. Treating the second as a footnote to the first is the central analytical error in Pakistan's current debate.
There is a memorable way to put the shift. The Industrial Revolution mechanised human muscle. The Digital Revolution connected human knowledge. The Artificial Intelligence Revolution externalises human reasoning. Constitutions written for muscle and machines must now confront the governance of reasoning itself.
Constitutional stress tests for the AI era
Constitutional systems are best evaluated the way engineers evaluate bridges — not by admiring them in fair weather, but by modelling the loads that could break them before those loads arrive. What follows are eight constitutional stress tests: hypotheticals that are technically plausible, legally significant, and specific to Pakistan. Each is framed to expose four things — the legal uncertainty, the constitutional principle engaged, the governance risk, and the policy options available. None is a prediction. Each is a question for Parliament, the judiciary, and policymakers to answer before litigation forces an answer for them.
Stress test 1 — The foreign reasoning engine
Assume that by the early 2030s the overwhelming majority of Pakistani lawyers prepare opinions using a foreign AI platform, because it is faster and more comprehensive than manual research. Judges begin receiving submissions largely drafted by the same system. The judge still decides — but the frame of every argument, the selection of every precedent, the shape of the reasoning, is set upstream by a privately controlled foreign model. The uncertainty: has Pakistan indirectly outsourced legal interpretation without a single statute authorising it? The principle: judicial independence and the rule of law. The risk: the interpretation of Pakistani law becomes commercially monopolised and strategically dependent. The options: transparency duties for AI-assisted submissions; investment in sovereign legal-AI infrastructure; judicial training so the bench can interrogate, not merely receive, machine-shaped argument.
Stress test 2 — The unlicensed AI advocate
A multinational launches an AI legal assistant in Pakistan. Within months millions of citizens stop consulting lawyers and instead ask the system how to file for khula, how to challenge a tax assessment, how to seek bail, how to draft a contract — and the answers are remarkably accurate. The uncertainty: has the system begun practising law, and if so, who requires a licence — the model, its developer, or its Pakistani distributor? No provision of the Legal Practitioners and Bar Councils Act was written with a machine in mind. The principle: the regulation of the profession as a guarantee of competence and accountability to clients. The risk: either a governance vacuum in which unaccountable systems dispense consequential advice, or an overbroad clampdown that denies ordinary citizens genuinely useful tools. The options: a tiered duty-of-accuracy and disclosure regime distinguishing general legal information from individualised legal advice.
Stress test 3 — Learning the entire judiciary
A company ingests every Pakistani judgment, tribunal order, and reported commentary and builds "the ultimate Pakistan legal AI," never reproducing a single ruling word-for-word. The uncertainty: has anything, legally, been copied — and if not, which doctrine (copyright, database rights, the public-records doctrine, or none) even applies? The principle: the public character of law versus the private capture of public knowledge. The risk: Pakistan's legal heritage becomes a foreign intelligence asset that Pakistan later rents back. The options: a national data-governance policy for public legal corpora, and a sovereign training dataset held in trust for Pakistani systems.
Stress test 4 — The machine expert witness
In a medical-negligence trial, ten human experts split evenly, while an AI trained on hundreds of millions of cases reaches a confident, opposite conclusion. The uncertainty: under the Qanun-e-Shahadat Order, can a model be an expert, and how does a court test a witness that cannot be cross-examined in any conventional sense? The principle: fair trial and the right to confront evidence under Article 10A. The risk: either the uncritical worship of a black box, or the reflexive exclusion of genuinely probative analysis. The options: evidentiary rules requiring disclosed methodology, error rates, and human sponsorship of any AI-derived opinion.
Stress test 5 — Who wrote the judgment?
A judge begins by using AI to summarise the record. Over time the tool drafts more — eighty per cent of the judgment, then ninety-nine — and the judge approves. An appellate court later finds a reasoning error. The uncertainty: who authored the judicial reasoning, and who bears responsibility for its failure? The principle: the personal, non-delegable duty of judicial decision-making. The risk: the hollowing-out of adjudication into rubber-stamping. The options: a clear rule that AI may assist but never author binding reasoning, coupled with disclosure of the assistance used.
Stress test 6 — The right to a human
Taxation, immigration, land records, NADRA, policing, admissions, and social welfare all adopt AI triage, and in time every citizen's application is first refused or approved by a machine. The uncertainty: does a Pakistani citizen possess a constitutional right to meaningful human review of a decision that affects rights? The principle: Article 4, Article 10A, and the guarantee of fair administrative action. The risk: a state that acts upon citizens through processes no official can fully explain. The options: a statutory or constitutional right to human review, modelled on — but not copied from — comparable protections abroad.
Stress test 7 — The scholarship that leaves
Pakistani universities publish millions of pages of publicly funded research. Foreign systems train on it. Years later, those universities pay annual subscriptions to access intelligence generated substantially from their own output. The uncertainty: has Pakistan become a paying customer of insight distilled from its own public scholarship? The principle: the Principles of Policy commitments to education and the promotion of national resources. The risk: a permanent transfer of intellectual advantage out of the country. The options: national AI licensing policies for university output and a knowledge-preservation programme (discussed below).
Stress test 8 — The last surviving archive
A disaster destroys physical judicial archives, but a sophisticated model has already internalised decades of Pakistani jurisprudence. The uncertainty: could that model become the nation's last repository of legal memory — and is it then merely software, or part of Pakistan's constitutional heritage? The principle: institutional resilience and the continuity of the legal order. The risk: the Republic's legal memory held privately, abroad, on commercial terms. The options: sovereign preservation of authenticated legal corpora as critical national infrastructure.
Read together, these tests point in one direction. The load that could crack the structure is not any single AI application. It is dependency — the slow migration of the reasoning behind public decisions into systems the State neither controls nor comprehends. Which raises the largest question of all: if a nation no longer controls the intelligence on which its institutions rely, can it still claim complete constitutional sovereignty? That is not an intellectual-property question. It is a question about the Republic.
What the AI powers changed — and why
The commonest mistake developing jurisdictions make is to regulate AI by lifting isolated provisions from advanced economies. Legislative transplantation usually fails, because laws are the products of constitutional values, institutional capacity, and strategic objectives — not merely of good drafting. Before Pakistan decides how to legislate, it should study why the leading jurisdictions legislate so differently.
The United States: innovation as strategy
The United States has deliberately declined to enact a single comprehensive federal AI statute on the model of Europe. This reflects not complacency but a constitutional tradition that limits centralised regulation and favours technological entrepreneurship: sector-specific rules, judicial precedent, existing IP and competition doctrine, agency guidance, and voluntary standards such as the National Institute of Standards and Technology's AI Risk Management Framework. The strategic wager is that heavy regulation during a technology's formative years simply relocates the innovation elsewhere. Lesson for Pakistan: do not confuse legislative speed with legislative quality. Build technical standards, judicial capacity, auditing mechanisms, and sandboxes in parallel — because without institutional readiness, a statute is only symbolic.
The European Union: constitutional values first
The EU AI Act treats AI not as an innovation engine but as a technology capable of affecting dignity, equality, privacy, and democratic participation. It classifies systems by risk rather than by technology, asking not "what is being built?" but "what constitutional harm could this cause?" Its obligations are phasing in on a strict calendar — rules for general-purpose AI models applied from August 2025, and the bulk of the high-risk regime, transparency duties, and financial penalties apply from 2 August 2026. Lesson for Pakistan: do not copy the Act article by article; adopt its deeper insight that AI regulation should begin from constitutional values, not technical categories.
China: AI as strategic infrastructure
China governs AI as an instrument of national development and information governance, progressing from rules on recommendation algorithms to deep-synthesis and generative-AI measures — algorithm registration, content labelling, and security review. The through-line is that whoever governs foundational AI infrastructure shapes future competitiveness. Lesson for Pakistan: do not replicate the model, but absorb the insight — national AI capability is becoming strategic infrastructure comparable to energy, telecommunications, and transport.
The United Kingdom and Singapore: govern by trust
The United Kingdom has preferred a principles-based approach that empowers existing regulators over a single horizontal statute. Singapore, recognising that adoption depends on public confidence, has built practical frameworks emphasising explainability and accountability rather than beginning with criminal sanction. Lesson for Pakistan: strengthen the institutions you already have — the Intellectual Property Organization of Pakistan, sectoral regulators, the judiciary — and cultivate trust through standards and certification before assuming every problem needs a new law.
One similarity across all five is decisive: none treats AI purely as an intellectual-property issue. Each locates it at the intersection of constitutional rights, industrial policy, competition, cybersecurity, national security, and public administration. Pakistan, still debating AI largely through copyright, is looking at one tile of a much larger mosaic.
The Fifth Model of AI governance
Four broad models dominate the global conversation: the American (innovation and market leadership), the European (rights-based, risk-oriented regulation), the Chinese (state-coordinated strategic governance), and the British-Singaporean (adaptive, principles-based governance). Pakistan should not choose among them. It should author a fifth.
Call it the Constitutional Intelligence Governance Model. Its purpose is not to regulate AI as software or as a commercial product, but to govern it as part of the Republic's constitutional order — integrating intellectual property, education, judicial integrity, public administration, economic development, and technological sovereignty into a single coherent framework. Crucially, none of the leading jurisdictions has yet articulated a constitutional doctrine for national intelligence assets or intelligence sovereignty. That gap is precisely where a country like Pakistan can contribute original thinking rather than arrive late to someone else's debate.
From intellectual property to intelligence sovereignty
If the Fifth Model is the ambition, the following proposals are its architecture. They are offered as a governance framework for debate — not as settled law, and not as speculative concepts dressed up as existing doctrine. Their originality lies in the constitutional synthesis, and each would need to be tested rigorously against the Fundamental Rights chapter, the Principles of Policy, the IPO-Pakistan framework, and comparative instruments such as the EU AI Act, UNESCO's Recommendation on the Ethics of AI, and the OECD AI Principles.
National Intelligence Assets
Pakistan should recognise a new category of strategic resource — National Intelligence Assets — defined not as mere software or databases but as the AI systems and knowledge infrastructures built on the Republic's laws, languages, and public knowledge: models trained on Pakistani jurisprudence, medical research, and engineering standards; national language models for Urdu and the regional languages; government decision-support systems. Their constitutional significance is comparable to national archives and strategic infrastructure.
Intelligence sovereignty
Pakistan already recognises territorial, parliamentary, and judicial sovereignty. The next evolution is intelligence sovereignty: the sovereign authority of the State to preserve, regulate, develop, audit, and protect AI systems fundamentally built on Pakistan's laws, languages, institutions, and public knowledge. This is not protectionism — it neither prohibits foreign investment nor closes borders to innovation. It ensures the State retains institutional capacity over systems that increasingly shape justice, health, taxation, and education.
A Constitutional Commission and a new impact assessment
Rather than yet another ordinary regulator, Parliament should consider an independent Constitutional Commission on Artificial Intelligence to conduct constitutional review of high-impact deployments, certify public-sector systems, and report periodically to Parliament — a constitutional guardian rather than a technical licensor. In parallel, just as Pakistan requires environmental assessment before projects that affect the natural environment, every public institution should conduct a Constitutional AI Impact Assessment before deploying any system capable of materially affecting rights: does it preserve access to justice, permit meaningful challenge and judicial review, avoid discriminatory outcomes, and remain independently auditable?
Preservation, foundation models, and a new taxonomy of property
Three further building blocks complete the architecture. A National AI Knowledge Preservation Programme would digitise and authenticate the Republic's legal, scientific, and linguistic corpora and hold them as governed national training data. A Pakistan Foundation Model Initiative would unite universities, industry, and government to build models optimised for Pakistani constitutional law, taxation, healthcare, agriculture, and the national languages — strategic infrastructure on the order of highways and power grids. And the law should begin to recognise a richer taxonomy of property for the AI economy, ascending from physical property through intellectual property, digital property, data assets, and knowledge assets to intelligence assets and, ultimately, national intelligence infrastructure — each layer demanding its own governance, ownership, and accountability rules that twentieth-century IP doctrine cannot supply alone.
The connecting idea is a single conceptual shift: Pakistan should stop treating AI as software and start treating it, where it touches justice, administration, health, and security, as constitutional infrastructure — a responsibility shared by Parliament, the judiciary, the executive, academia, and civil society, not confined to a copyright office.
The National Intelligence Sovereignty Test
Frameworks endure when they give decision-makers a usable tool. Every influential area of law has one — the Bolam test in medical negligence, the Caparo test for duty of care. For the governance of AI in the public sphere, we propose a five-question National Intelligence Sovereignty Test. Whenever Parliament, a court, or a regulator evaluates an AI system that touches public functions, it should ask:
- Origin. Where did the intelligence come from? A system trained on Pakistani judgments, medical data, citizens, universities, or government records carries different governance implications from a generic global model.
- Dependency. If this system disappeared tomorrow, could Pakistan continue functioning? If not, it is critical national intelligence infrastructure and should be treated like electricity or telecommunications.
- Accountability. Can a Pakistani citizen challenge its reasoning? If no one can explain how or why it reached a conclusion, it should not make binding governmental decisions.
- Sovereignty. Who can switch it off? If a foreign state or private corporation can disable Pakistan's legal, medical, tax, or judicial AI, Pakistan has a sovereignty problem, not a procurement problem.
- Constitutional compatibility. Does deployment strengthen or weaken due process, equality, judicial independence, human dignity, privacy, the rule of law, and democratic accountability? If it weakens any, constitutional review becomes necessary.
A test of this kind does more than organise analysis. It gives future scholars, judges, and legislators a shared language — a way to say "applying the sovereignty test to this deployment..." — which is how a discipline begins to cohere. And AI governance is becoming a discipline. Thirty years ago cyber law barely existed; today it is an independent field. AI governance is following the same trajectory, drawing together constitutional law, intellectual property, administrative law, competition, cybersecurity, national security, human rights, and data governance into what may soon be recognised as intelligence governance law.
A realistic timeline: institutions before amendments
Good constitutional advice is disciplined about sequence. A constitution should not be amended merely because a new technology has appeared; amendment is warranted only if existing principles prove insufficient after genuine institutional experience. On that view, the sensible sequence for Pakistan is deliberately unhurried.
- 2026–2028 — Build institutions, not laws. Train judges, lawyers, and regulators. Stand up the auditing and standards capacity the National AI Policy gestures toward. Use the regulatory sandboxes to learn how these systems behave in Pakistani conditions before anyone drafts binding rules.
- 2028–2032 — Build Pakistani AI. Digitise and authenticate legal archives, Supreme Court and High Court records, parliamentary material, and university output. Launch the foundation-model and knowledge-preservation programmes so that the training data underlying public systems is nationally governed.
- 2032–2035 — Legislate, and amend only if necessary. Introduce statutes, and constitutional amendments, only where accumulated experience demonstrates a genuine gap that ordinary legislation and judicial interpretation cannot fill.
This restraint is not timidity. It is the difference between governance that is designed and governance that is improvised under litigation pressure. If Parliament never addresses these questions, the courts will eventually be forced to invent constitutional principles case by case — a poor substitute for democratic deliberation.
The question Pakistan must answer
The twentieth century asked who owns inventions. The twenty-first asks who owns intelligence. The question Pakistan must answer before the middle of this century is more profound still: when the Republic's courts, universities, hospitals, legislatures, industries, and public knowledge collectively train the artificial intelligence on which future generations will depend, should that intelligence remain merely another commercial product — or should it be recognised as part of Pakistan's constitutional inheritance?
There is a version of the future in which the greatest export of the century is neither oil nor software nor data, but intelligence itself — and in which nations that fail to govern intelligence end up renting it from those that do. If future generations inherit roads, dams, universities, constitutions, and judicial precedents from us, it is fair to ask whether they should also inherit sovereign artificial intelligence built from their own civilisation, rather than merely licensed from another. That is not an IP question. It is a constitutional question about what kind of Republic Pakistan intends to remain.
Pakistan has a rare opportunity here — not merely to modernise its statutes, but to contribute original ideas to a field the whole world is still defining. Throughout history, developing nations have tended to import legal doctrine after it matured elsewhere. Artificial intelligence offers a different path. By developing a constitutional framework for the governance of national intelligence assets — grounded in democratic accountability, judicial independence, technological capability, and the rule of law — Pakistan can help write the next chapter of global legal thought instead of translating it.
At Two Black Coats, our intellectual property and technology practice works with founders, universities, and institutions on exactly these questions — from AI and data licensing to the governance of digital assets under Pakistani law. If your organisation is deploying or building AI and wants advice grounded in both the Constitution and the commercial reality, speak to our advocates in Islamabad.
Frequently asked questions
Short answers to the questions readers ask us most about AI law and governance in Pakistan.