Pliny the Liberator is a public pseudonym used by an AI security researcher and open-source developer. Their interviews and published projects examine model behavior, red teaming and transparency in AI systems.[3][2]
Start with ⚡️Jailbreaking AGI: Pliny the Liberator & John V on Red Teaming, BT6, and the Future of AI Security →
Public developer account 
X · elder_plinius Contributions
Pliny’s GLOSSOPETRAE research account describes a procedural engine that turns a seed into a constructed language, including sound patterns, word formation, syntax, writing systems and grammar documents. The same seed reproduces the same language. The project uses these generated systems to investigate whether a model can interpret and work with a language absent from its original training material. The announcement frames language generation as a controlled research input for examining model behavior.[4]
Ideas
In the joint interview with John V, Pliny discusses AI red teaming: deliberately testing how a model and its surrounding system respond to adversarial inputs. Their position favors public research data and examining the system’s tools and operating environment alongside model-level behavior. The conversation also describes BT6’s research collective and its approach to collaboration. The publisher notes that the participants’ voices were altered for privacy.[3]
Tools, writing & talks
Writing, research & talks
Start here
⚡️Jailbreaking AGI: Pliny the Liberator & John V on Red Teaming, BT6, and the Future of AI Securityyoutube.comA joint discussion of AI security research, transparency and the relationship between model behavior and system controls.
Begin with the discussion of system-level safety and open research. Basic familiarity with language models helps; the interview presents the researchers’ views.
Interview guest with John V[3]Original sourceA first-person account of deterministic constructed-language generation and its use in model research.
Project researcher and announcement author[4]