What Nobody Is Telling Pakistan About “Recovery”
A dramatic recovery figure can create a false impression if its components are not separated. If land worth Rs1 trillion is restored to the state, that is a major institutional recovery, but it does not create Rs1 trillion of immediately spendable fiscal space. The state may still need to secure possession, remove encroachments, update revenue records, defend the title in court, determine permissible land use and prevent the same property from being captured again.
A credible public recovery dashboard should therefore distinguish between:
| Recovery category | Public disclosure required |
|---|---|
| Cash deposited | Amount, receiving account and deposit date |
| Assets returned to victims | Number of victims, distribution status and unclaimed balance |
| Land mutation | Area, location, former allotment, lawful owner and mutation record |
| Physical possession | Date possession was secured and department taking custody |
| Court-dependent recovery | Case number, legal status and appeal exposure |
| Valuation | Valuation date, methodology and responsible authority |
| Indirect recovery | Exact conduct prevented and basis for assigning monetary value |
Without this separation, Pakistanis cannot tell how much was actually received, how much remains tied up in litigation and how much represents estimated protection of an asset from loss. Transparency here is not anti-NAB. It is the strongest defence NAB could build against accusations that headline valuations are being substituted for measurable outcomes.
The same governance principle applies to AI-personalised financial reporting: technology may reorganise facts and surface patterns, but it must never alter the underlying record or conceal the reasoning behind a decision. Likewise, as explained in the discussion on workplace artificial intelligence, human responsibility cannot disappear simply because an algorithm participates in the workflow.
The Safeguards NAB’s AI System Will Need
Every AI-generated lead should remain a lead until a trained investigator corroborates it. Every important recommendation should carry an audit trail showing the source records used, the system version, the authorised user and the human decision that followed. Original evidence must remain preserved separately from AI-generated summaries or interpretations.
Access should be role-based and independently logged. Models should be tested for false positives across languages, regions, naming conventions and transaction patterns. Defendants must be able to challenge evidence produced through the system, while courts should receive a clear explanation of whether AI discovered a relationship, summarised documents or materially influenced an investigative conclusion.
Most importantly, an institution must never generate a prosecution narrative first and use AI to search backwards for supporting fragments. That would automate confirmation bias, not accountability.
Pakistan’s advantage is that this system is reportedly being developed with the National Centre of Artificial Intelligence as an indigenous platform, providing an opportunity to build data sovereignty and institutional ownership from the beginning. The danger is assuming that “sovereign” also means automatically secure, accurate or impartial. Those properties must be engineered, tested and audited.
Frequently Asked Questions
Did NAB recover Rs5.41 trillion in cash?
No. NAB describes the figure as direct and indirect recoveries, with much of the value arising from recovered state land and other assets. Rs42.303 billion was specifically reported as money and assets returned to fraud victims.
Is NAB’s AI investigation system already fully deployed?
Reporting in June described the system as established and demonstrated, with completion and deployment still being advanced. NAB’s August statement says the bureau is shifting towards an AI-based e-investigation regime. Its precise nationwide deployment status and operational coverage have not been publicly itemised.
