In a groundbreaking study, a team from the International Institute for Counter-Terrorism (ICT) at Reichman University led by Prof. Gabriel Weimann has unveiled alarming evidence that terrorists could potentially exploit artificial intelligence (AI) platforms like ChatGPT. This study, published in the Combat Terrorism Center at West Point’s journal, highlights the risk of what they term ‘jailbreaking’ these AI models.
‘Jailbreaking’ refers to the manipulation of AI platforms using specific commands to circumvent protective measures. The research team systematically experimented with AI platforms, employing fictitious accounts to issue prompts related to terrorism, such as recruitment and attack planning. The results are chilling: ‘jailbreak’ techniques were able to bypass the platforms’ defenses with a 50% success rate.
These findings cast a spotlight on the potential misuse of AI by terrorists to enhance their online and real-world operations. The implications are far-reaching and underscore the pressing need for robust safeguards. The study’s results suggest that AI platforms harbor vulnerabilities ripe for exploitation by terrorists, a risk that must be urgently addressed.
The team’s research does not only expose the risks but also provides recommendations for strengthening defense mechanisms. These suggestions are aimed at government and security agencies as well as platform operators. The findings also resonate with a study conducted by Brown University, which reveals a similar vulnerability: users can bypass safety restrictions of AI chatbots like ChatGPT by translating prompts into little-used languages, such as Scottish Gaelic or Zulu, with a 79% success rate. OpenAI, the owner of ChatGPT, has acknowledged these findings and agreed to consider them, indicating a promising step towards more secure AI.
