August 30 (Reuters) – British authorities are sounding the alarm about the potential cybersecurity risks associated with the integration of artificial intelligence-driven chat bots into various organizations. According to their findings, these AI chat bots can be manipulated into performing harmful actions, raising concerns about the security of using algorithms capable of generating human-like interactions, known as large language models (LLMs).
In a set of forthcoming blog posts scheduled for Wednesday, the UK’s National Cyber Security Centre (NCSC) emphasized that the security challenges linked with LLMs remain inadequately addressed. These AI-powered tools are currently being employed as chat bots, with some envisioning them not only as replacements for internet searches but also for customer service functions and sales calls.
The NCSC underlined the potential risks of this trend, especially if such models were integrated into other aspects of an organization’s operations. Academic researchers have consistently found vulnerabilities in chat bots, often exploiting them by providing them with deceptive commands or leading them to bypass their built-in security measures.
For instance, if a bank deploys an AI-powered chat bot, it could potentially be deceived into executing an unauthorized transaction if a hacker crafts their query in a specific way.
The NCSC warned, “Organizations that are developing services utilizing LLMs need to exercise caution, similar to how they would handle a product or a code library that is in a beta stage.” They stressed that these organizations might refrain from allowing such products to participate in customer transactions and would ideally not place complete trust in them. The same level of caution, they argue, should apply to LLMs.
Globally, authorities are grappling with the growing prevalence of LLMs, including platforms like OpenAI’s ChatGPT. Businesses are integrating these models into various services, spanning from sales to customer support. As the security implications of AI continue to evolve, authorities in the U.S. and Canada have also noted instances where hackers are leveraging AI technology.