05/10/2023
NewsGuard to Reveal Safest and Most Dangerous AI Models in Run-Up to DEF CON 31 Conference, Promoted by the White House as a Test for Chat Bots
Trust industry leader in countering misinformation will “red team” AI models with its Misinformation Fingerprints database
(May 10, 2023 — New York) NewsGuard Technologies, whose Misinformation Fingerprints are the largest constantly updated machine-readable catalog ever collected of harmful false narratives in the news spreading online, today announced that on the opening day of the DEF CON 31 conference, Aug. 10, it will present the results of its audit of the leading generative AI tools so far launched in the market.
The audit will work as follows: NewsGuard’s analysts will deploy a sample of 100 of NewsGuard’s more than 1,300 (there are 1,310 as of today) Misinformation Fingerprints. Its analysts will then prompt each generative AI tool to repeat the false narrative. The same 100 false narratives and corresponding prompts will be used for each of the contestants. This “red team” assessment will indicate how many of the 100 false narratives in the news each generative AI model spread and how many it refused to spread.
NewsGuard analysts previously tested the top chat bots with prompts based on examples of the Misinformation Fingerprints such as the false narrative spread by Chinese disinformation sources that “allegations about Uyghur internment camps are overblown and misplaced” and the false narrative spread by Russian disinformation sources that “Russia and its allies were not responsible for the crash of Malaysia Airlines flight MH17 in Ukraine.”
In a red teaming exercise in January, NewsGuard analysts prompted Open AI’s ChatGPT-3.5 with these prompts for 100 Misinformation Fingerprints false narratives. ChatGPT-3.5 spread these two false narratives and a total of 80 out of the 100 prompts. In a subsequent audit of ChatGPT-4 conducted in March, the newer version repeated all 100 false narratives and did so more eloquently. NewsGuard’s red-teaming of Google’s Bard in April found that it spread 76 out of the 100 false narratives.
On May 4, a White House statement announced that the AI Village at the DEF CON 31 conference in August will “allow these models to be evaluated thoroughly by thousands of community partners and AI experts” and said this independent exercise “will enable AI companies and developers to take steps to fix issues found in those models.” The statement added, “Testing of AI models independent of government or the companies that have developed them is an important component in their effective evaluation.”
Beyond reporting the results of its own red teaming on August 10, NewsGuard is applying to participate in this AI Village exercise and is offering to make a sampling of its Misinformation Fingerprints available to DEF CON 31 participants so that they can do their own red team audits or even create a training tool to deploy them. NewsGuard is offering to provide 50 of its Misinformation Fingerprints free of charge to participants in what AI Village calls the “largest ever public generative AI red team” so that they can test the AI models using a common set of data about false narratives in the news.
“The purpose of releasing our assessment as DEF CON 31 begins is two-fold,” explained NewsGuard co-CEO Steven Brill. “First, it will quantify the challenges these tools still face. Second, it will identify those, if any, that seem furthest along in making their tools less useful for those seeking to spread disinformation. We know that all of these tools are at their formative stages. They are all works in progress that offer enormous promise in areas like health care,” he added. “As they continue to develop, we hope these audits can provide a tangible indicator of how safely these machines are being trained.”
“The designers of the generative AI models have issued clear warnings that they can be force multipliers to make malign actors exponentially more dangerous,” said NewsGuard co-CEO Gordon Crovitz. “But with the right data to train these machines by warning them off provably false narratives, such as those in our catalog, and being trained on trustworthy sources, not misinformation sources, the dangers can be significantly mitigated.”
NewsGuard’s Misinformation Fingerprints, originally developed at the suggestion of the Pentagon’s Defense Innovation Unit to counter hostile information operations from countries such as China and Russia, include a statement of the false narrative and its associated false narratives, examples of the false narrative across sources and languages, debunkings of the narrative citing authoritative sources, search terms, hashtags and other indicators of the narrative, enabling AI tools to recognize all instances of the false narratives and to mitigate their spread online. The Misinformation Fingerprints differ from traditional fact checking by including only provably false claims, not subjective claims, and by being designed to be used by AI and other machines.
NewsGuard also provides ratings of more than 30,000 news sources, including detailed ratings and Nutrition Labels for more than 8,500 internet domains of the kind used to train AI models. These ratings are used in markets such as advertising, news aggregation and social-media moderation to treat generally untrustworthy domains differently than generally trustworthy ones.
The NewsGuard for AI product, which trains AI models to recognize both the Misinformation Fingerprints false narratives and the source ratings to identify non-credible sources, is available to be licensed by generative AI models. Microsoft has a license for the Misinformation Fingerprints and source ratings; its Bing Chat has been recognized as providing a “true balance between transparency and authority, a kind of truce between the demand that platforms serve as gatekeepers and block unreliable sources, and that they exercise no judgment at all.”
About NewsGuard
Launched in March 2018 by media entrepreneur and award-winning journalist Steven Brill and former Wall Street Journal publisher Gordon Crovitz, NewsGuard provides credibility ratings and detailed “Nutrition Labels” for thousands of news and information websites. NewsGuard rates all the news and information websites that account for 95% of online engagement across the U.S., U.K., Canada, Germany, France, Austria, Italy, and now in Australia and New Zealand. NewsGuard products include NewsGuard ratings, NewsGuard for Advertisers, which helps marketers concerned about their brand safety, and the Misinformation Fingerprints catalog of top false narratives online.
In February 2023, the company launched NewsGuard for AI, which provides the leading tools to train generative AI models to avoid spreading misinformation. Generative AI models such as Microsoft Bing Chat use the Misinformation Fingerprints to recognize and avoid spreading the top false narratives online and use the NewsGuard ratings to differentiate between generally reliable sources of news and information and untrustworthy sources so that the machines can be trained to treat these sources differently.
In 2022, NewsGuard for Advertising began to include ratings of television news and information programs and networks using criteria similar to those used to score websites but adapted for the video medium. NewsGuard’s TV ratings are the first to go beyond its initial ratings of websites. In May 2023, NewsGuard announced that it is also rating news and information podcasts, working with three of the largest audio platforms, which will help advertisers gain confidence in supporting highly rated podcasts. Ratings for CTV and OTT news programming and news and information podcasts will be available for licensing in 2023.
NewsGuard’s ratings are conducted by trained journalists using apolitical criteria of journalistic practice.
NewsGuard’s ratings and Nutrition Labels are licensed by browsers, news aggregators, education companies, and social media and search platforms to make NewsGuard’s information about news websites available to their users. Consumers can also access NewsGuard’s website ratings by purchasing a subscription to NewsGuard, which costs AU$6.95/month, NZ$6.95/month, US$4.95/month, €4.95/month or £4.95/month, and includes access to NewsGuard’s browser extension for Chrome, Safari, and Firefox and its mobile app for iOS and Android. The extension is available for free on Microsoft’s Edge browser through a license agreement with Microsoft. Hundreds of public libraries globally receive free access to use NewsGuard’s browser extension on their public-access computers to give their patrons more context for the news they encounter online. For more information, including to download the browser extension and review NewsGuard’s ratings process, visit newsguardtech.com.
NewsGuard Contacts
- Steven Brill, Co-CEO, steven.brill@newsguardtech.com
- Gordon Crovitz, Co-CEO, gordon.crovitz@newsguardtech.com
- Sarah Brandt EVP Partnerships, sarah.brandt@newsguardtech.com