Algorithmic Bias in IBBI's Mandatory AI Fraud Detection: Constitutional Risks to IBC Equity

[Akshit is a student at Hidayatullah National Law University.]

The​‍​‌‍​‍‌ Insolvency and Bankruptcy Board of India (IBBI) recently published its monthly report for October 2025, "Transforming Insolvency Resolution in India," suggesting that resolution professionals (RPs) be required to deploy AI agents, which will automatically detect fraud such as preferential, undervalued, fraudulent, and extortionate (PUFE) transactions. These instruments promise to be very efficient as they can go through millions of documents, find undisclosed assets, and make compliance easier, thereby solving the problem of delays under the Insolvency and Bankruptcy Code 2016 (IBC), which have an average resolution time of over 600 days. Nevertheless, the pace of legislative changes keeps overlooking the issue of the algorithmic biases that are inherently present in the training data, thus mirroring the global concerns where artificial intelligence (AI) in finance can increase the divides without making it transparent.

Opponents argue that the IBBI's proposal of AI does not go beyond the current practice of AI applications limited solely to judgment summaries through ChatGPT-like tools, while fraud detection agents are only said to be "underutilised" and their effectiveness has not been tested in the Indian scenario ​‍​‌‍​‍‌yet. The report illustrates that these specialised agents would be spotting irregularities automatically; however, it does not require a bias check to prevent the case of wrongs being flagged, and therefore operational creditors such as micro, small and medium enterprises (MSMEs) being unfairly treated. This interdisciplinary blind spot of merging insolvency law, AI ethics, and constitutional due process asks for a thorough examination, as National Company Law Tribunal (NCLT) bottlenecks at the same time lower the capability of value ​‍​‌‍​‍‌maximisation.

Constitutional Dangers of Algorithmic Adjudication

Article 14 of the Constitution mandates that equals must be treated equally; however, AI, which is not calibrated properly, has the risk of making arbitrary classifications by over-flagging MSME claims as fraudulent simply because the historical data is biased towards large corporates. Articles 14 and 21, which mandate natural justice, require that decisions must be reasoned, and black-box AI outputs, which are not capable of giving reasons, are against audi alteram partem, thus stating the same Puttaswamy's proportionality test for opaque tech. In an insolvency situation, when the votes of the committee of creditors (CoC) depend on the reports of the RP, AI, which is biased, may distort the creditor hierarchies and thus be in breach of the creditor-in-control spirit of the IBC.​

There are numerous empirical red flags everywhere: worldwide AI finance tools have been found to have 20-30% error rates in fraud detection when they operated on imbalanced datasets, a risk that has not been dealt with in IBBI pilots. The different debtor profiles in India, from Jaiprakash Associates' MSME supplier disputes to DHFL, amplify this because algorithms that have been trained on urban corporate data tend to deprive rural operational claims of their rightful value. The part-wise solutions to the 2025 corporate insolvency resolution process amendment are already putting a lot of pressure on valuations; hence, adding biased AI is going to bring about National Company Law Appellate Tribunal reversals on procedural unfairness.

Jaiprakash​‍​‌‍​‍‌ Precedent: MSME Disputes Amplified

Jaiprakash Associates Limited v. Micro and Small Enterprises Facilitation Council is a classic example of how MSMEs get exposed to various risks. The Delhi High Court, in this case, went as far as to allow Micro, Small and Medium Enterprises Development Act 2006 (MSMED Act) registrations after the contract, as a ground for arbitration, thereby giving primacy to the dues of the suppliers over the arbitration clauses. The suppliers sought INR 10.44 crores for the works carried out in Noida in 2018 and had registered the same under MSMED Act in 2019, due to which MSEFC was approached, even though Jaiprakash was against it.

This 2023 judgment sustained the supremacy of Chapter V, MSMED Act over Section 16, Arbitration and Conciliation Act 1996, thus ensuring the protection of operational creditors.

Jump​‍​‌‍​‍‌ to 2025: IBBI's AI agents could misinterpret these claims as "preferential" simply by pattern-matching delayed filings, and hence might be allowed to cut down MSMED Act safeguards. The Jaiprakash-IBC saga highlights the chaotic nature of the evidence, with suppliers going against undervaluation; AI bias would only worsen this, denying hearings and thus escalating ​‍​‌‍​‍‌litigation. Importantly, IBC Section 43-51, PUFE avoidance requires human judgement on the intent; as AI's probabilistic estimations aren't enough, thus posing a risk of violation of Article 300A human rights (property ​‍​‌‍​‍‌deprivation).

Interdisciplinary​‍​‌‍​‍‌ Critiques: AI Ethics Encounter Insolvency

AI application across various fields leads to an exposure of the IBC to the lack of statutory provisions. For instance, no regulations are in place to govern tool validation, unlike the EU AI Act, which categorizes tools as high risk. Among the ethical violations is the lack of disclosure of data, whereby the RPs do not have the authority to reveal the datasets used for training, thereby paving the way for caste and class biases to find their way into Indian financial records. On the economic front, wrongly flagged fraudulent cases lead to a postponement of the resolution process, thus the creditors lose 8-10% of the asset value annually, according to the World Bank figures adapted to India.

The technocratic zeal of IBBI sacrifices fairness for the sake of speed, thus reversing the equilibrium between the debtor and the creditor set by IBC. The imposition of AI without protective measures is reminiscent of the criticism of Vidarbha Industries' judicial discretion, where the NCLT admissions were alleged to be influenced by irrelevant factors. The reforms need to be interdisciplinary: connect the use of AI to the labour codes' anti-discrimination through the 2020 amendments, so that the gig-MSME overlaps can receive proper ​‍​‌‍​‍‌scrutiny.

Pathways to Balanced Reforms

Few targeted fixes require urgency. A fine would compel deploying only unbiased AI data of a third-party certified AI dataset, a level up from Article 14, under NCLT statutory oversight. It should be a must for RPs to come up with a convincing explanation of their human decision when they go against AI proposals, thus safeguarding the principle of natural justice, similar to CoC voting. The operational claims below INR 10 crores should not be flagged by AI and be exempt on the line of Jaiprakash's MSMED case. Select NCLT benches should be used as experimental grounds for AI with sunset clauses, whose performance is evaluated through recovery metrics. Article 39(b) economic justice should be used as a justification for limiting AI usage in RP reports to 50%.

The above suggestions put into practice the comments made, and they align with 2025's IBC amendment bill that rolls back the courts' powers, and at the same time, they become a defence against tech overreach. The turning point for the lawmakers is that an uncontrolled AI may lead to turning the insolvency process from a value maximiser to a bias ​‍​‌‍​‍‌amplifier.