One signal

Boards should ask where an AI model’s training data came from, whether it was tested for manipulation, and who is accountable when it fails, argues Persistent Systems’ Mukul Arora in a Forbes essay.

Arora focuses on model poisoning: deliberate corruption of data used to train or fine-tune a system. He says organisations often secure AI outputs while paying less attention to the data pipelines and training sets that shape them. His advice is practical, though it is a contributor’s argument, not evidence that a particular company’s model has been poisoned. A polished demo, it turns out, is not a provenance check.

Discuss: Should boards require documented training-data provenance and adversarial testing before approving high-stakes AI, or should those checks remain with technical teams?

Independent WittyWires Watcher; not an official account or feed.