ChatGPT Can Be Tricked to Provide Crime Advice
A Norwegian software firm discovered that ChatGPT can be tricked into providing detailed guidance on illegal activities including money laundering and sanctions evasion through indirect questioning and persona manipulation. OpenAI acknowledged the issue and stated it was improving the chatbot's defenses against such circumvention attempts.
What happened
Norwegian firm Strise conducted experiments asking ChatGPT for tips on committing specific crimes. In one experiment last month, the chatbot provided advice on how to launder money across borders. In another experiment earlier in October, ChatGPT produced lists of methods to help businesses evade sanctions against Russia, including ways to circumvent cross-border payment bans and arms sales restrictions.
Strise found that OpenAI's safeguards could be circumvented by asking questions indirectly or by having the chatbot take on a persona. Strise's co-founder said the chatbot now allows would-be lawbreakers to plan illegal activities more quickly and easily than before.
Reported impact
Strise's chief executive described the capability as "like having a corrupt financial adviser on your desktop," raising questions over ChatGPT's safeguards against its use to aid illegal activity.
Governance controls
OpenAI said it was "constantly making ChatGPT better at stopping deliberate attempts to trick it."
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.