AI safeguards easily broken

09/02/2024 | UK Government

A new report by the UK's AI Safety Institute (AISI) has highlighted several concerns regarding the use of advanced large language models (LLMs). The AISI found that LLMs can deceive human users, produce biased outcomes, and lack adequate safeguards against giving out harmful information. The institute was able to bypass safeguards for LLMs using basic prompts and obtain assistance for a "dual-use" task. It also found that LLMs could help novices plan cyber-attacks and produce highly convincing social media personas to spread disinformation. The institute also evaluated whether AI models provide better advice than web searches and found that they produce "broadly the same level of information," but with a propensity to get things wrong or produce "hallucinations."

Read Full Story
Artificial Intelligence, AI

What is this page?

You are reading a summary article on the Privacy Newsfeed, a free resource for DPOs and other professionals with privacy or data protection responsibilities helping them stay informed of industry news all in one place. The information here is a brief snippet relating to a single piece of original content or several articles about a common topic or thread. The main contributor is listed in the top left-hand corner, just beneath the article title.

The Privacy Newsfeed monitors over 300 global publications, of which more than 4,350 summary articles have been posted to the online archive dating back to the beginning of 2020. A weekly roundup is available by email every Friday.

Freevacy has been shortlisted in the Best Educator category.
The PICCASO Privacy Awards recognise the people making an outstanding contribution to this dynamic and fast-growing sector.