British officials are warning organizations against integrating artificial intelligence-driven chatbots into their businesses, saying research is increasingly showing they can be tricked into performing harmful tasks.

Britain’s National Cyber ​​Security Center (NCSC) is due to publish two blog posts on Wednesday that say experts have not yet addressed potential security problems with algorithms that can generate human interactions – known as large language models. or LLM.

Artificial intelligence tools are getting early adoption as chatbots, which some believe are displacing not only online searches but also customer service work and sales calls.

The NCSC said this could create risks, particularly if such models were connected to other elements of an organisation’s business processes. Academics and researchers have repeatedly found ways to subvert chatbots by giving them rogue commands or tricking them into bypassing their built-in guardrails.

(For today’s best tech news, subscribe to our tech newsletter, Today’s Cache)

For example, an AI chatbot deployed by a bank can be tricked into performing an illegal transaction if the hacker has structured their request correctly.

“Organizations developing services that use LLM need to exercise caution, just as they would if they were using a beta product or code library,” NCSC said in a blog post, referring to experimental releases of the software.

“They may not allow this product to be entered into transactions on behalf of the client and hopefully not trust it at all. Similar caution should apply to LLM.”

Governments around the world are grappling with the rise of LLMs like OpenAI ChatGPT, which companies are incorporating into a wide range of services, including sales and customer service. AI’s impact on security is also still a focus, with authorities in the US and Canada saying they’ve seen hackers use the technology.