Anthropic has released its report on the misuse of Claude for the month of September. While various instances of misuse were detected, one case that caught our attention involved its use in Malaysia to manipulate the state elections held this year.
Claude was used to disseminate information intended to campaign for specific parties. Over a thousand fake accounts on X were utilized for this purpose, alongside the creation of the "Malaysian Pulse" website, where all content was generated using artificial intelligence (AI).
These fake accounts were found to be disseminating identical content in a coordinated manner, garnering over a million views on social media. Claude was employed to target voters using electoral district profiles constructed from actual census and voter data. The AI-generated content was then tailored based on race, religion, and attitudes toward the royal institution.
Following Anthropic's detection of the activity, the entire operation was tracked to prevent future misuse. Claude rejected several instructions from the operators, even when attempts were made to bypass safeguards using different prompts. Anthropic has since blocked the IP addresses and account names used by the operators.
