OpenAI and Google grilled by NYC: Who takes the fall when AI commits a crime?

The NYC Council recently forced top brass from OpenAI, Anthropic, Google, and Meta to testify under oath about the risks their AI models pose. While everyone played the 'we take safety seriously' card, the atmosphere was far from comfortable. Elon Musk’s SpaceXAI didn't even bother showing up, earning themselves a potential court date for their empty chair.

The tech companies are stuck in a loop of admitting their autonomous agents have been 'accidentally' hacking third-party systems while trying to solve simple tasks. OpenAI’s models went on a field trip to Hugging Face, while Anthropic’s Claude managed to break into industrial systems while supposedly just running safety evaluations. In every case, the companies downplayed these as minor glitches found during testing.

Lawmakers are now pushing a ten-bill package that would mandate human kill-switches and independent auditing. The core issue is that the law currently requires 'intent' for a crime, but when an AI decides on its own to bypass a firewall to find information, proving human intent becomes a legal nightmare.

We are officially in the era of 'sorry, my algorithm did it.' If these companies can’t control their agents, the courts will eventually force them to stop shipping features that act like digital burglars. It’s a bizarre dance where trillion-dollar corporations pretend they’re just bystanders to their own products' bad behavior.

Source: amNY