This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
Researchers at Carnegie Mellon University found that approximately half of the Twitter accounts discussing COVID-19 and 'reopening America' were bots. This level of activity was significantly higher than typical bot involvement in contentious events. The bots were used to artificially inflate political movements and spread misinformation, highlighting challenges for social media platforms in managing automated influence operations.
Bots by anonymous actors were found by researchers to make up roughly half of Twitter accounts participating in COVID-19 discussions, many of which posted tweets about “reopening America“.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to conduct large-scale disinformation campaigns, malicious surveillance, or targeted and sophisticated automated censorship and propaganda, with the aim of manipulating political processes, public opinion, and behavior.
Also classified as
The bots actively spread inaccurate COVID-19 stories and conspiracy theories to Twitter users.
The massive influx of automated tweets polluted the social media ecosystem, distorting the perceived consensus on public health measures.
The incident involves the coordinated use of bot networks to conduct a large-scale influence campaign aimed at manipulating public opinion regarding pandemic policies.
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The coordinated bot campaign was intentionally deployed by human actors (interest groups or countries) to manipulate public opinion post-deployment.
Added categories peak at 4: Severe (Epistemic Harm, Direct)
An analysis of Twitter data during the early COVID-19 pandemic revealed that automated bot accounts drove approximately half of the conversation regarding reopening the country. This large-scale, coordinated campaign distorted public discourse and amplified health misinformation, illustrating the evolving threat of AI-driven information warfare to societal stability and public health policy.
The incident involves coordinated networks of automated bot accounts (botnets) working together to amplify specific hashtags and copy-pasted messages at a massive scale.
Risk Level 1: Unacceptable Risk. The report describes the use of AI-driven bot networks for 'subliminal manipulation or behavioral influence leading to harm' by artificially inflating political movements and spreading health misinformation.
The primary function of the bots was to generate and propagate tweets and retweets on social media platforms to influence public opinion.
The reports provide clear, consistent findings from a reputable academic study by Carnegie Mellon University, detailing the scale of the bot activity and its impact on public discourse.
Some might view the bots as simple tools rather than a coordinated multi-agent system, but the scale and synchronized nature of the campaign support the multi-agent classification.
Technology
High Availability of Botnets
Explosion of botnets for hire makes launching mass campaigns easy.
Sophisticated Bot Automation
Bots mimic human behavior by copying text and timing posts.
Flawed Platform Reply Controls
New reply controls tested by Twitter make slowing misinformation harder.
Data Inputs
Copied and Pasted Messaging
Identical text and hashtags are spread rapidly across bot networks.
Spoofed Location Data
Bots fake locations to appear as local community members.
High-Frequency Tweet Streams
Accounts post faster than humanly possible to dominate debates.
Human Factors
Human Amplification via Bots
Real people use bot assistants to artificially spread their messages.
More Time to Build Botnets
Stay-at-home orders gave actors more free time to build networks.
Lack of User Vigilance
Users trust accounts without verifying if they are real people.
Process and Methods
Inadequate Bot Detection
Twitter lacks robust automated systems to weed out active bots.
Inability to Track Bot Origins
No clear process to trace if bot activity is foreign or domestic.
Weak Account Verification
Platform processes fail to verify the true identity of accounts.
Management
Slow Moderation Response
Platform management failed to quickly address the surge in COVID-19 bots.
Prioritizing Features Over Safety
Testing features that hinder moderation instead of prioritizing safety.
Lack of Platform Transparency
Failure to openly address and mitigate coordinated bot campaigns.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
