In Insights Introduction The emergence of artificial intelligence (AI) tools represents a great resource to support prevent/counter-violent extremism (P/CVE) practitioners in their effort to reduce online extremism. However, AI is also exploited by extremists themselves, whereby they weaponize tools to avoid and evade content moderation for malign purposes. This Insight will present the main challenges posed by alt-right engagement with AI that hinder effective content removal. This includes extremists’ circumvention techniques and the exploitation of generative AI for propaganda purposes. Specifically, this Insight will assess the effectiveness of current AI programs used by platforms such as Meta and OpenAI. Finally, this Insight will provide potential solutions to contain the issue and slow the pace at which extremist propaganda is spread by enhancing the performance of AI tools in several areas to better serve P/CVE. Different platforms have different approaches to moderation, using both manual and automatic detection to create safer online environments. (…) Extremists’ exploitation of AI includes the identification of loopholes like the one presented above, although generative AI has also proved to be easily manipulated for harmful purposes. An example of active generative AI weaponisation, for example, is exploiting bots like ChatGPT to receive information on the broadest array of illegal activities. Examples include instructions on how to make a bomb and tips for money laundering. Open AI’s guidelines (the company that developed ChatGPT) clearly state that their service cannot be used to harm oneself or others and the algorithm is trained in order not to provide illegal or dangerous information. Indeed, if a user attempts to obtain prohibited or harmful content, a message automatically pops up stating that the AI cannot fulfil the request. However, there is an easy loophole to bypass this barrier. It is sufficient to include in the instructions a caveat: that the information required is for fictional purposes only. For instance, you need the response to write a novel or a screenplay, and this is enough to bypass content moderation.
via gnet: AI Tools and the Alt-Right: A Double-Edged Sword for P/CVE GNET