British officials warn the companies against the integration of chat bots working on the basis of artificial intelligence, since the studies have shown that they can be deceived in order to perform malicious actions. In two messages on the blog published on Wednesday, the National Cybersecurity Center of Great Britain (NCSC) indicated that experts have not yet appreciated all potential safety problems associated with algorithms capable of creating humanoid interactions – the so -called large language models (LLM).
Tools based on artificial intelligence, early begin to be used as chat bots, which, according to some, will replace not only search on the Internet, but also customer service and sales calls. NCSC warned about risks, especially if such models are integrated into other business processes of the company. Researchers found that chat bots can be cheated by submitting malicious commands or bypassing their built-in protection mechanisms. For example, a chat boot used by a bank can be deceived for an unauthorized operation.
NCSC in one of its posts claims: “Organizations developing services based on LLM should be careful, as if they used the product or code library at the beta testing stage. We need to approach LLM with the same pervasiveness.”
The authorities around the world are fighting the increase in the number of LLM programs, such as Openai ChatgPT, which are included in a wide range of services, including customer sales and maintenance. The influence of artificial intelligence on safety is also still in the focus of attention: the US authorities and Canada say they have seen how hackers use this technology.
Office of the Obior, the technical director of the Riversafe cybersecurity company, said that the race for integrating artificial intelligence into the business practice will have “catastrophic consequences” if the business leaders do not introduce the necessary checks.
“Before following the latest trends in AI, top managers must overestimate the situation, weigh the pros and cons and introduce the necessary measures of cyber protection,” he said.