INVESTSIGHT CAPITAL is a fintech and capital markets firm in Bengaluru, India. This article is part of our research library - for the official site see investsightcapital.com, or ask us a question.
An automated workflow may process more information than a human can read. That capability leaves another question unresolved: who is responsible when its output is wrong or inappropriate?
In a February 2026 address, SEBI’s chairman emphasised human accountability, privacy and fairness in the use of AI. Read the original address. A speech provides context; it should not be represented as a product approval or a new rulebook.
Define the boundary of the task
State whether the tool organises documents, calculates a defined measure or suggests material for review. An output labelled as a research aid should not silently become an instruction to transact.
For example, a document classifier can route a company announcement to an analyst. The analyst still needs to read it, establish its significance and decide whether the research note changes.
Identify the review owner
Assign someone to check consequential outputs, document corrections and escalate unclear cases. Include a fallback for absence or a service interruption. A workflow that depends on an unspecified reviewer has a gap even if its interface looks complete.
Measure failures that matter
Track missing evidence, incorrect periods, omitted documents and unresolved exceptions. A polished demonstration may hide these problems because it only shows the cases selected to work.
Good responsibility statements explain what the system does, what it cannot establish and how a person can question its result. They do not promise that human review removes every risk.
