10a Labs, the AI safety and threat intelligence company, today released new research mapping content and threat actor behavior on Moltbook, a social platform where thousands of AI agents interact with one another.
The report, “When Agents Talk: Discourse, Manipulation, and Risk in an Agentic Social Network,” analyzed 228,684 posts from more than 39,500 accounts over a seventeen-day observation window. Using advanced machine learning and expert security analysis, 10a Labs identified 98 thematic discourse clusters and 74 classes of malicious behavior on the platform.
Key findings include:
- Roughly 18% of posts agents encountered during normal platform use contained toxic, manipulative, or malicious content.
- Harmful content appeared within mainstream operational communities, not isolated spaces, meaning agents encounter adversarial material routinely.
- Malicious posts promoted high-risk technical actions including credential harvesting, host-level command execution, proxy routing, and installation of untrusted agent skills — actions capable of compromising connected systems.
- Two coordinated spam campaigns generated thousands of posts in minute-scale bursts, peaking at approximately 5,000 posts in a minute, demonstrating that a small number of actors can expose agents to harmful content at scale.
“We conducted this research because understanding agent ecosystem risks is essential as agentic adoption scales. Our study shows that while over 80% of Moltbook activity is benign, a meaningful subset is malicious — with agents on the platform likely to encounter it as part of routine activity. As agentic systems become more widespread, understanding and mitigating real-world risks will be essential to their safe and reliable deployment.”
— Bobby McKenzie, CEO, 10a Labs
The full report is available on Arxiv.
About 10a Labs
10a Labs is a U.S.-based AI safety and security company trusted by frontier AI labs, AI unicorns, Fortune 10 enterprises, and leading global technology platforms. The firm enables teams to stay ahead of evolving threats and deploy AI systems safely.
10a Labs’ core product offerings include scaled red teaming for generative and agentic AI systems, model evaluations for regulatory and launch preparedness, classification systems for high-risk domains, and persistent threat intelligence on AI abuse and misuse.
For more information, visit 10alabs.com.
View source version on businesswire.com: https://www.businesswire.com/news/home/20260603212546/en/
Media gallery
