Compliance with the Digital Private Knowledge Safety (DPDP) Act alone won’t be sufficient for banks as they scale synthetic intelligence (AI), Reserve Financial institution of India (RBI) Governor Sanjay Malhotra stated, whereas discussing six key dangers from the know-how – explainability, bias and exclusion, focus and herding, third occasion and vendor dependence, information privateness and cyber and adversarial vulnerabilities.
Talking on the FIBAC 2026 convention in Mumbai on Tuesday, Malhotra stated AI has vital potential to rework banking and monetary inclusion, however its deployment have to be accompanied by robust governance, accountability and human oversight.
“The Reserve Financial institution sees it really as a functionality to be responsibly harnessed, and never merely as a threat to be contained,” he stated.
Malhotra stated the dangers related to AI should be addressed at the same time as banks transfer sooner on adoption, significantly given the know-how’s potential to change the economics of credit score supply, enhance customer support, scale back operational prices and strengthen fraud prevention.
Six dangers from AI
The primary threat Governor spoke about was the black field or explainability downside.
He stated banks want to have the ability to clarify selections made by AI, significantly when these selections materially have an effect on prospects.
“If a mortgage software is rejected, why was it rejected?” he requested, including that auditors, prospects and regulators would all need to know the explanation behind such selections.
“Auditors want to know, prospects want to know, regulators want to know,” Malhotra stated.
The second threat is bias and exclusion.
He warned that algorithms skilled on historic lending information may be taught and perpetuate present biases, together with these linked to geography, occupations or communities.
“We now have to make sure that equity in AI isn’t a compliance check-box; it’s a design requirement from day one,” he stated.
The third threat is focus and herding, which may emerge if monetary fashions developed by a small variety of know-how distributors grow to be extensively used throughout the banking system.
Monetary establishments counting on related fashions for credit score and buying and selling selections may face system-wide penalties if a mannequin incorporates an error or bias, Malhotra stated.
The fourth threat is third-party threat and vendor dependence.
The Governor acknowledged that banks, significantly smaller establishments, could have to depend on exterior know-how suppliers as a result of they might not have the sources to construct their very own giant language fashions and AI capabilities.
Nonetheless, he stated such dependence must be actively managed.
“Compliance can not simply cease at our outer partitions,” Malhotra stated.
The fifth threat is information privateness and compliance with the DPDP Act. Whereas banks have to adjust to the laws, Malhotra stated merely assembly the statutory requirement shouldn’t be thought of enough.
“We can not stay glad or justify simply complying with it; the truth is, we have to do way more,” he stated.
The sixth threat is cyber and adversarial vulnerability. Malhotra said that AI techniques themselves can grow to be a supply of cyber threat, citing experiences from July by which an AI system was reportedly in a position to collaborate on testing and attacking techniques.
“These AIs themselves can assault you,” he stated, including that banks should be significantly cautious in regards to the new vulnerabilities created by more and more succesful AI techniques. Financial institution, not algorithm, stays accountable
Past the six particular dangers, Malhotra stated essentially the most basic concern is the potential erosion of human judgment and accountability.
