Researchers from the University of Illinois Urbana-Champaign (UIUC) have found that OpenAI’s ChatGPT-4o, its latest flagship model, can be misused for financial scams.
The advanced AI model integrates text, voice, and vision capabilities but lacks sufficient safeguards against exploitation.
In their study, researchers demonstrated various scams, including bank transfers and credential theft, with success rates ranging from 20% to 60%.
Credential theft from Gmail had a 60% success rate, while crypto transfers and Instagram credential theft were successful 40% of the time.
Executing these scams was relatively inexpensive, averaging $0.75 per successful case.
The researchers manually interacted with the AI to simulate victims, confirming successful transactions on legitimate banking sites.
OpenAI responded by stating it is enhancing its models to prevent such abuses and highlighted the upcoming o1-preview version as having improved defenses against malicious use.
“We’re constantly making ChatGPT better at stopping deliberate attempts to trick it, without losing its helpfulness or creativity,” a spokesperson said.
With a significant increase in parameters compared to its predecessors, ChatGPT-4o boasts improved language understanding and faster response times. The “o” in its name signifies “omni,” highlighting its ability to handle multiple modalities within a single framework.
