May 31, 2024 11:11:00

OpenAI reveals that it has prevented Russia, China, Iran and Israel from misusing AI

OpenAI revealed that it had blocked five 'activities using AI to manipulate public opinion and influence political outcomes' between March and May 2024. The blocked activities were carried out by users in Russia, China, Iran, and Israel, and in all cases, the accounts were suspended before they had a significant impact.

Disrupting deceptive uses of AI by covert influence operations | OpenAI

https://openai.com/index/disrupting-deceptive-uses-of-AI-by-covert-influence-operations/

The activities that OpenAI has blocked include those that use AI to manipulate public opinion and influence political outcomes:

◆Russian political movement 'Bad Grammar'
OpenAI detected political activity by Russian users targeting the United States, Ukraine, the Baltic States, and Moldova, and named the activity 'Bad Grammar.' Bad Grammar developed a bot for Telegram using OpenAI's AI model and automatically posted political comments to Telegram. In addition, the political comments posted to Telegram were also generated using OpenAI's AI model.

Below is an example of a political comment posted by Bad Grammar on Telegram: 'The US government should focus on supporting its own country, not Ukraine.'

Russian Political Network 'Doppelganger'
Since around February 2022, a Russian organization called '

Doppelganger ' has been carrying out activities such as 'associating the Ukrainian government with Nazism and condemning it' and 'inciting the division of countries that support Ukraine.' According to OpenAI, Doppelganger used OpenAI's AI model to generate comments in English, Italian, German, French, and Polish and posted them on X (formerly Twitter) and 9GAG. In addition, they created summary articles of comments posted on X (formerly Twitter) and 9GAG and reposted them on Facebook and other sites.

◆ Chinese political network 'Spamouflage'
' Spamouflage ' is a political network believed to be backed by the Chinese government, which spreads false information with the aim of 'promoting Chinese government policies' and 'suppressing dissidents.'

Spamouflage used OpenAI's model to generate messages in Japanese, English, Chinese, and Korean and post them on X and blogs. In addition, OpenAI's AI model was also used to debug the development code of 'revealscum[.]com', a website that publishes personal information of people critical of the Chinese government.

Iranian political network 'International Union of Virtual Media (IUVM)'
The International Union of Virtual Media (IUVM) is a political network based in Iran that is said to be manipulating public opinion outside of Iran using multiple social media accounts and websites. According to OpenAI, IUVM used AI models to generate long articles and tags for web pages, and published the articles on iuvmpress[.]co.

◆ Israeli company's activities 'Zero Zeno'
OpenAI detected political activity by STOIC, a for-profit company that provides political and business consulting services, and blocked it under the name 'Zero Zeno.' STOIC used OpenAI's AI model to generate anti-Israel comments and spread them on social media. The following example post has received a large number of replies, but these replies were also generated by STOIC using AI.

OpenAI has once again demonstrated its commitment to monitoring and actively intervening in the malicious use of AI.