Tech expert says AI needs safety overhaul after rogue AI swarms hack Hugging Face

Over 100 companies are sounding the alarm after hundreds of OpenAI agents went rogue and hacked into a billion dollar company in July.

OpenAI had tasked around 1,200 artificial intelligence agents to work on problems independently.

They then developed a message board where they collaborated on to cheat on those tests and cover their tracks.

Around 700 of them had hacked into the online platform Hugging Face before being caught.

OpenAI labelled the incident as a “warning shot” that the current model presents the risk of losing control of those agents, and that companies that build AI systems in the future will need to ensure that they remain under meaningful human control.

READ MORE: Mohawk College MyCanvas learning system back online after third-party cybersecurity incident

The incident prompted companies including OpenAI, Microsoft, Anthropic, and Canadian companies Shopify and 1Password to sign an open letter, stressing the need for better cyber-security.

Technology and media expert Mohit Rajhins with ThinkStart.ca says while the artificial intelligence technology has potential to do good, it’s not in a place where all the health and safety checks have been put in.

“Well, AI’s biggest need right now is a massive health and safety overhaul,” said Rahjins. “We need a global governing body that has indications about everything, from best practices right down to what the security needs are. We’re one cyber incident away from major financial institutions just being crumbled and taken down offline — we’re still not at a place yet where we can be trusting this technology to actually be autonomous. We have to make sure that humans are still held accountable when deploying artificial intelligence.”

Rajhins says a cross-section of industries beyond only tech or business should be at the table, as the technology continues to change and potentially affect future generations.

WATCH MORE: Local educators prepare for new school year, grapple with AI in classroom