John V

John V, who also publishes as John Vaina, works in AI red teaming and prompt engineering. He co-founded Bossy Discord and helps steer the hacker collective BT6’s commitment to open-source security data. His work also extends to AI evaluations, validation, and risk assessment, including policy work concerning AGI and nuclear strategic stability.[2][3][1][4]

With a background in prompt engineering and computer vision, John helped build Bossy Discord and shape BT6’s open-data ethos. In its December 16, 2025 interview with John and Pliny the Liberator, Latent Space summarized that ethos as “if you can’t open-source the data, we’re not interested.”[1]

Latent Space reported that John and Pliny together turned down enterprise gigs and pushed back on Anthropic’s closed bounties. Their shared position placed AI security at the system layer, rather than relying on restrictions within a model’s latent space. These decisions made access to security findings a condition of the work they would take on.[1]

For its February 20, 2026 AI red-teaming session, LimaCharlie described John as working in AI risk, safety, and security at the Institute for Security and Technology (IST). The organizer described his work as spanning adversarial machine learning, AI evaluations and validation, and risk assessment, alongside policy work at the intersection of AGI and nuclear strategic stability.[4]

IST’s AI-NC3 working-group roster, published June 1, 2026 and updated August 13, lists John in AI Technical Development and Trajectories, alongside Pliny the Elder and other experts. The initiative’s working groups develop policy recommendations for integrating AI into nuclear command, control, and communications systems.[5]

Before these 2025–2026 security and policy appearances, John had published a concrete prompt-engineering proposal. It illustrates the structured prompting side of his technical work.[3][1][4]

John published Polyrhythmic-Reasoning Prompt Pattern on June 20, 2024; the article was subsequently revised in April 2026. Its protocol pairs a four-item sequence addressing technical, conceptual, practical, and theoretical problems with a five-item sequence addressing confused, inquisitive, hurried, skeptical, and undecided users. Both proceed through logical deduction, evaluation, and expansion. For example, a skeptical user prompts evidence or examples, followed by evaluation of that evidence and expansion with more sources.[3]

The musical analogy supplies an output rule: repeat the four-item sequence five times and the five-item sequence four times, then “Only display results when both sequences align every 20 steps.” John presented better reasoning, efficiency, compute savings, and usability as goals. These are proposed benefits of the prompt instructions, rather than benchmarked results or a demonstrated account of a model’s internal reasoning.[3]