When Elon Musk's xAI was forced to apologise this week after its Grok chatbot spewed antisemitic content and white nationalist talking points, the response felt depressingly familiar: suspend the service, issue an apology and promise to do better. Rinse and repeat.

This isn't the first time we've seen this playbook. Microsoft's Tay chatbot disaster in 2016 followed a similar pattern. The fact that we're here again, nearly a decade later, suggests the AI industry has learnt remarkably little from its mistakes.

But the world is no longer willing to accept 'sorry' as sufficient. This is because AI has become a force multiplier for content generation and dissemination, and the time-to-impact has shrunk. Thus, liability and punitive actions are being discussed.

The Grok incident revealed a troubling aspect of how AI companies approach accountability. According to xAI, the problematic behaviour emerged after they tweaked their system to allow more 'politically incorrect' responses - a decision that seems reckless. When the inevitable happened, they blamed deprecated code that should have been removed. If you're building systems capable of reaching millions of users, shouldn't you know what code is running in production?

The real problem isn't technical - it's philosophical. Too many AI companies treat bias and harmful content as unfortunate side effects to be addressed after deployment, rather than fundamental risks to be prevented beforehand. This reactive approach worked when the stakes were lower, but AI systems now operate at unprecedented scale and influence. When a chatbot generates hate speech, it's not embarrassing - it's dangerous, legitimising and amplifying extremist ideologies to vast audiences.

Link: https://economictimes.indiatimes.com/opinion/et-commentary/rogue-bots-ai-firms-must-pay-up/articleshow/122525366.cms?from=mdr