By Raphael Satter
(Reuters) – British officials are warning organisations about integrating artificial intelligence-driven chatbots into their businesses, saying that research has increasingly shown that they can be tricked into performing harmful tasks.
In a pair of blog posts due to be published Wednesday, Britain’s National Cyber Security Centre (NCSC) said that experts had not yet got to grips with the potential security problems tied to algorithms that can generate human-sounding interactions – dubbed large language models, or LLMs.
The AI-powered tools are seeing early use as chatbots that some envision displacing not just internet searches but also customer service work and sales calls.
The NCSC said that could carry risks, particularly if such models were plugged into other elements organisation’s business processes. Academics and researchers have repeatedly found ways to subvert chatbots by feeding them rogue commands or fool them into circumventing their own built-in guardrails.
For example, an AI-powered chatbot deployed by a bank might be tricked into making an unauthorized transaction if a hacker structured their query just right.
“Organisations building services that use LLMs need to be careful, in the same way they would be if they were using a product or code library that was in beta,” the NCSC said in one its blog posts, referring to experimental software releases.
“They might not let that product be involved in making transactions on the customer’s behalf, and hopefully wouldn’t fully trust it. Similar caution should apply to LLMs.”
Authorities across the world are grappling with the rise of LLMs, such as OpenAI’s ChatGPT, which businesses are incorporating into a wide range of services, including sales and customer care. The security implications of AI are also still coming into focus, with authorities in the U.S. and Canada saying they have seen hackers embrace the technology.
(Reporting by Raphael Satter; Editing by Alex Richardson)
