SEBI Deploys AI Surveillance Against Misleading Finfluencer Content

CNBC-TV18 reports that SEBI Chairman Tuhin Kanta Pandey described the survey result as a reason for greater digital oversight. "This digital vigilance is further necessitated by our latest Investor Survey, which revealed that 62% of investors are influenced by finfluencers - many of whom operate without accountability or verified performance data," Pandey said, according to the outlet.

The reported detection scope includes guaranteed-return claims and fraudulent certifications. India Today notes that misleading investment promotion does not necessarily appear as plain text: videos, images, advertisements and messages can all carry investment claims. The approach extends beyond text-only monitoring by considering several content formats and contextual signals together.

SEBI's 2025-26 annual report describes a wider technology program that includes data analytics, AI and advanced surveillance tools. The report states that the regulator is using these technologies to strengthen market oversight, improve risk preparedness against cyber threats and support investigation mechanisms.

SEBI has paired Project SUDARSAN with R(AI)DAR, another AI-enabled system. CNBC-TV18 reports that R(AI)DAR reviews advertisements issued by asset management companies and is used to identify misleading or unauthorised promotional material.

The two systems address distinct monitoring surfaces:

• •Project SUDARSAN focuses on unsolicited financial advice and suspicious public digital activity, according to India Today and CNBC-TV18.
• •R(AI)DAR focuses on advertisements from asset management companies, CNBC-TV18 reports.
• •Both tools form part of SEBI's use of technology and data analytics in supervisory work, as outlined in the regulator's annual report.

The available reporting does not specify the underlying model architecture, training data, accuracy metrics, false-positive rates or the threshold that triggers regulatory review. It also does not establish that a system-generated alert by itself results in an enforcement action.

The public description illustrates a high-stakes content-risk workflow: multimodal ingestion, regional-language analysis, risk scoring and human examination of structured alerts. In comparable regulatory deployments, model quality depends not only on detection coverage but also on auditable review processes, calibrated thresholds and clear evidence trails for downstream investigators.

The technical challenge is especially relevant in financial-content monitoring, where a legitimate market discussion, promotional claim and prohibited or deceptive recommendation can share overlapping vocabulary. Industry experience with comparable systems indicates that multilingual context, image and video interpretation, and reviewer feedback loops are central to reducing both missed harmful content and erroneous flags. SEBI has not publicly disclosed operational performance measures for Project SUDARSAN in the material reviewed.