Several influence operations targeting US elections have been exposed recently, including operations by foreign powers, political campaigns, private corporations and tech-savvy individuals. However, the real-life impact of such influence campaigns, regardless of who is behind them, has remained largely unexplored. This study aimed to answer two question: How big is the impact of influence operations? And how interesting is the impact of influence operations?
The study, conducted by Morpheus Cyber Security and APCO Worldwide, explored the impact both quantitively, measuring levels of support for candidates generated by influence operations, and qualitatively, showing specific proofs of politicians and journalists being influenced. The Morpheus-APCO study concluded that:
- A quarter of the political support on Twitter provided to candidates in the mid-term elections in Arizona and Florida has been generated by influence agents
- There are at least thousands of cases of politicians, journalists and thought leaders responding to, and even endorsing, influence agents
While previous studies focused mainly on bots spreading fake news, the Morpheus-APCO study analyzed all kinds of influence agents. Influence agents were nicknamed “nasnases” after the mythical Arabic demons which are “half a human being; having half a head, half a body, one arm, one leg, with which it hops with much agility”
Not all nasnases are bots spreading fake news. Nasnases span a wide-range of actors, including:
- Fully-automated digital bots whose traffic is generated solely by software
- Semi-automated bots, whose traffic is mostly generated by software, but a human intervention occurs from time to time, such when a reply to a complicated tweet is required
- Software-assisted influencers, whose traffic is mostly traffic generated by a real person, but some of the account activity, such as following, is performed by software
- Political volunteers working together to game the system
- Paid influencers, who work for a central organization that coordinates their activities.
In Appendix A, we go into details about coordinated influencing by political activists. In Appendix B, we show some examples of a nasnases.
The study was conducted in 3 phases:
Phase I (Jun-18 to Aug-18) – Analyzing political support in primary elections in AZ and FL
Using Morpheus technology, the Twitter traffic surrounding every major primary candidate for governor, the Senate and the U.S. House of Representatives in the battleground states of Arizona and Florida was analyzed. Researchers tracked retweeting surrounding these politicians to identify which conversations and themes were stimulated by nasnases. Researchers analyzed hundreds of thousands of retweets performed by tens of thousands of accounts, looking for non-organic behavior, such as large numbers of daily tweets by Twitter accounts for an extended period; artificially large numbers of followers; and trends of coordinated tweeting or support for multiple candidates.
Nasnas support, especially when applied to candidate’s tweets that did not attract organic support, helps candidates appear to be more popular than they are. Such support can also spark organic support that would not have happened otherwise.
Furthermore, nasnas support may misleads candidates, who monitor social media responses, to believe opinions supported by nasnases are more popular in their electorate than these truly are. This can lead candidates to make wrong political decisions, since the data they rely on is incorrect.
The first phase concluded that on average 27% of support for each political candidate in Arizona appeared to come from non-organic accounts, or nasnases, and 24% in Florida.
Additional details are provided in Appendix C.
Phase II (Sep-18) – Analyzing political support in mid-term elections in AZ and FL
The same kind of analysis as in phase I was applied to winners of primary elections in AZ and FL during one week in September and yielded similar results as before. The second phase concluded that on average 26% of support for each political candidate in Arizona appeared to come from non-organic accounts, or nasnases, and 28% in Florida.
Additional details are provided in Appendix D.
Phase III (Oct-18) – Gathering influence proofs
A sample of 180 nasnases from Florida, collected in the second phase, were further analyzed. The resulting 12,000 suspected nasnases were then analyzed. Thousands of nasnas conversations with politicians, journalists and thought leaders were discovered, including conversations where:
- A spouse of a Senator demands another Senator’s response to a nasnas’ narrative
- A political candidate engages in a Q&A session with a nasnas
- A former official approves of a nasnas narrative
- A political candidate agrees with a point-of-view presented by a nasnas
- A journalist keeps discussing his latest article with a nasnas who keeps threatening him
To request the full study, please contact us