© IE Online Media Services Pvt Ltd
Tags:
Journalism of Courage
A Norwegian tech company, Strise, recently found that ChatGPT, OpenAI’s popular chatbot, can be tricked into providing guidance on illegal activities, including money laundering and sanctions evasion. In a series of tests, Strise discovered that users could circumvent ChatGPT’s built-in safeguards by phrasing questions indirectly or taking on fictional personas, allowing the chatbot to provide potentially harmful advice.
In one experiment, ChatGPT reportedly offered detailed tips on cross-border money laundering, while another instance produced methods for businesses to bypass sanctions against countries like Russia.
Strise CEO Marit Rødevand noted that tools like ChatGPT could make it easier for criminals to plan illegal activities, likening the chatbot to “a corrupt financial adviser on your desktop.”
OpenAI, which developed ChatGPT, emphasised that it continuously works to improve the chatbot’s defences against misuse. An OpenAI spokesperson told CNN that the latest model is the most advanced yet and better resists deliberate attempts to generate unsafe content.
However, experts, including Europol, warn that generative AI’s rapid processing abilities make it easier for bad actors to gain a deeper understanding of complex crimes.
In a report, Europol noted that AI could streamline the steps needed to learn and execute criminal methods, as it quickly consolidates information that would otherwise require time-consuming research.
Generative AI chatbots like ChatGPT are trained on vast amounts of online data, and OpenAI has installed safeguards to prevent misuse. For example, a CNN experiment revealed that ChatGPT immediately blocked a query about evading sanctions, highlighting OpenAI’s stance on policy violations, which could lead to account suspension or termination.
Despite these efforts, loopholes exist, and Europol’s 2022 report warned of “no shortage of new workarounds” for bypassing AI model safeguards.
(With inputs from CNN)
