Justice Syed Mansoor Ali Shah of Pakistan’s Supreme Court entered a sentence into the court file in April 2025 that seemed almost too light for its weight. He said that artificial intelligence should be greeted with “careful optimism.” Ishfaq Ahmed versus Mushtaq Ahmed was a civil appeal, which is a type of procedural matter that typically goes unnoticed. Instead, it turned into something more akin to a constitutional moment: the first time a senior court in the area distinguished between artificial intelligence (AI) as a helpful tool and AI as a replacement for human judgment. This line, along with the global argument it represents, is now making its way to banking litigation with a growing sense of urgency.
When an AI system influences a financial decision—a loan denial, a fraud classification, or a credit assessment—and that decision is contested in court, who is responsible? This is essentially the same question that is circulating through courtrooms and legal departments from Islamabad to London. The algorithm is unable to defend itself. In cross-examination, the model is unable to defend itself. Furthermore, the entire proceeding is based on a fictitious argument if the legal research used to support the case was created by an AI tool that created a citation. This is not a theoretical issue. It is taking place.
| Detail | Information |
|---|---|
| Landmark judicial AI ruling | Pakistan Supreme Court (April 2025) — AI may support but never substitute judicial judgment · grounded in Article 10A (fair trial) and Article 37(d) (expeditious justice) |
| Justice Shah’s core principle | AI is “an assistant, not an authority” — opacity, bias, and hallucination identified as the three primary dangers to judicial integrity |
| AI hallucination risk | AI tools generating fictitious legal citations or invented case law — described as “corrosive” to any system based on binding precedent |
| US attorney-client privilege ruling | Federal court ruling (Feb 2026) — AI-generated documents may not satisfy attorney-client privilege; AI tools are not attorneys and do not hold confidential relationships |
| AI fraud in banking — key methods | Deepfake audio/video · voice cloning · automated conversational agents · highly personalised synthetic content generated at scale — often indistinguishable from human fraud |
| Legal framework gaps | Research (Journal of Banking Regulation, Jan 2026) identifies inconsistencies in UK’s principles-based framework — insufficient transparency, accountability, and legal remedies for AI-enabled financial fraud |
| International judicial AI standard | Council of Europe CEPEJ Ethical Charter (2018) — five principles: fundamental rights · non-discrimination · data quality · transparency · human oversight |
| Data protection concern | Court records contain confidential medical, family, and financial data — feeding such data into commercial AI platforms without legal safeguards raises serious jurisdictional and privacy risks |
| Recommended safeguards | Human review of all AI outputs · mandatory disclosure of AI use in judgments · verification of all AI-generated citations against authorised legal sources before use in decisions |
A US federal court decided in February 2026 that communications processed by public AI tools might not meet the requirements of attorney-client privilege because AI tools are not lawyers, do not maintain confidential relationships, and may lose the legal protection that allows for open legal advice. This decision carried significant weight for banking attorneys who had discreetly started utilizing AI to write memos, summarize discovery, and look up precedent. The improvements in efficiency were genuine. Apparently, the legal exposure was as well.
AI-enabled fraud is a parallel issue in financial institutions, and it is growing in ways that traditional anti-fraud systems were not intended to deal with. Fraudsters can create deepfake audio and video, alarmingly accurate voice clones, and customized communications that are nearly identical to authentic correspondence thanks to generative AI tools. Over the course of the first half of 2019 and the first half of 2024, UK banking fraud losses increased steadily.
The UK’s principles-based regulatory framework has real inconsistencies in how it addresses AI-driven fraud, according to research published in the Journal of Banking Regulation in January 2026. These inconsistencies include inadequate transparency requirements, ambiguous accountability chains, and legal remedies that were designed for a different type of threat. As the researchers point out, defenders are constantly catching up with evolving attack techniques.
Justice Shah’s framework, which views AI as a helper rather than an authority, provides a helpful framework for considering the future of banking litigation. He noted three risks unique to AI in legal settings: bias, hallucination, and opacity. The inability to trace how an AI system arrived at its output is known as opacity. In a legal context, this is not only a technical annoyance but also a constitutional flaw since an unexplained decision cannot be properly appealed or challenged.
Bias is more complex: if historical discrimination is encoded in the datasets used to train an AI model, those patterns may be replicated in the model’s outputs, potentially perpetuating structural disparities in lending, credit, and fraud classification that courts must subsequently resolve. Furthermore, for any system based on binding authority, hallucination—AI systems creating fake case citations or creating legal precedent—is genuinely destructive.
Observing these threads coming together in several jurisdictions at once gives the impression that the legal profession is taking its time facing an issue that it ought to have addressed sooner. Years ago, banks started incorporating AI into their decision-making processes, frequently more quickly than the regulatory frameworks governing those decisions were updated. Transparency, nondiscrimination, and required human oversight are among the five fundamental principles outlined in the Council of Europe’s Ethical Charter on AI in Judicial Systems, which was published back in 2018. There hasn’t been a consistent adoption of those principles. Additionally, there is currently conflict in courtrooms due to the discrepancy between the actual state of AI in financial and legal systems and what the regulations anticipated.
The exact shape of a final decision on AI decision tools in banking litigation is still unknown, as is the jurisdiction from which it will originate with sufficient power to influence international practice. Plaintiffs contesting AI-generated loan decisions, attorneys losing privilege claims over AI-assisted work products, and fraud victims coping with synthetic attacks their banks weren’t designed to recognize are all examples of the pressure that is increasing simultaneously. The algorithm is now present in the courtroom. Whether the law can keep up with it is the current question.
