Sophisticated Influence Campaign Powered by Claude: Multi-National Influence Disguised as Authentic Dialogue
Anthropic has uncovered an unusual and dangerous exploitation of its large language model, Claude, in a global campaign designed to influence public opinion via social media bots. More than 100 fake profiles operated simultaneously on Facebook and X (formerly Twitter), generating comments, shares, and discussions that appeared entirely human—but were actually authored and directed by AI.
The campaign aimed to promote seemingly moderate political narratives while supporting the interests of countries such as the United Arab Emirates, Iran, Kenya, and an unidentified European nation. These digital personas communicated in natural language, frequently employing humor and irony, and responded in real-time—not just based on content, but according to timing and emotional resonance.
Not Just Text: Coordination, Imagery, and Human-Like Response
Claude Managed the Operations—Just Like a Real Campaign Headquarters
What sets this campaign apart is not only the use of Claude to generate text but also its use for the strategic management of interactions: determining when to respond, in what tone, and on which platforms. The infrastructure behind the campaign included a well-programmed database managing names, political stances, interaction histories, and preferred topics. This created a consistent illusion of human activity, integrated with image generation tools and the ability to deflect “bot” accusations using humor and sarcasm.
Beyond Politics: Claude as a Tool for Cybercriminals
AI Shortens the Path to Advanced Crime
In addition to the influence campaign, Anthropic revealed other severe instances of Claude being exploited for criminal purposes. These include searching for security camera passwords, generating scripts for Brute Force attacks, developing sophisticated malware without requiring prior coding knowledge, and refining social engineering tactics targeting job seekers. Simply put, tools like Claude allow novice actors to leverage powerful models to create undetectable malware and execute advanced attacks in a short timeframe.
Recommendations from IPV Security Information Security Experts:
* Monitor suspicious activity of political personas on social media—particularly when they appear “too moderate” or excessively consistent.
* Implement controls on the use of internal AI tools, especially those with content generation or automation capabilities.
* Re-evaluate usage policies for language models—including clear definitions of permitted use, monitoring, and access controls.
* Train employees to identify bots and programmed dialogues, even within internal or professional groups.
* Prepare for automated influence campaigns, particularly surrounding elections, regulatory decisions, or strategic business moves.
In Conclusion
Anthropic’s disclosure highlights how easily artificial intelligence can become a powerful tool for manipulation. As AI adoption expands, there is an increasing need for regulatory frameworks, monitoring, and a reassessment of ethics and control. The digital world is changing rapidly—and it is our responsibility to lead that change rather than be driven by it.
For more information: https://thehackernews.com/2025/05/claude-ai-exploited-to-operate-100-fake.html