The quick spread of Generative AI (GenAI) tech has security experts worried. A new Lakera survey shows 95% of cyber security experts don’t trust current safety measures. Even though around 1,000 experts explored GenAI, only 5% think the safety steps are in place and can protect these AI systems well enough.
Lakera’s CEO, David Haber, pointed out how easy it is to break into GenAI models. “Even a beginner can trick AI systems with a few clever words leading to mistakes and data leaks,” he said. These vulnerabilities show the new dangers GenAI brings where old cyber defences often don’t work.
Lakera’s AI game, Gandalf, shows how big this problem is. Over a million people have played it, including cybersecurity experts. What’s worrying is that 200,000 players beat all seven levels. They got the AI to do things it wasn’t supposed to. On average, these players needed only 45 minutes to fool the system at level seven. This proves how easy it is to take advantage of GenAI models.
Joe Sullivan, who served as CSO at Cloudflare, Uber, and Meta and now advises Lakera, emphasized the need to tackle these security issues. “The rush to embrace GenAI driven by demands from the C-suite makes security readiness more crucial now than at any key point in tech’s history,” Sullivan explained. “To tap into its possibilities, companies must think about its hurdles, with security risk at the top of the list.”
Worries about LLM (Large Language Model) dependability and precision, data protection, and the shortage of trained staff pose major obstacles to GenAI adoption. Even with these obstacles, 22% of companies have used AI-specific threat modeling to get ready for GenAI-specific risks.
As GenAI keeps shaking up industries, companies need to step up their game when it comes to security. There’s a lot on the line, and businesses have to make it a top priority to address these vulnerabilities to keep their work and information safe.
