Definition
Plain language
A Brazilian AI lab whose work appears in cross-model persuasion studies.
As stated in the literature
A Brazilian AI research company that, with JusBrasil, published the cross-model persuasion-based guardrail-override study.
Why it matters: Regional AI labs like Maritaca expand the set of voices and threat models being studied beyond the dominant U.S.-and-U.K. centers.
For example, Maritaca AI partnered with JusBrasil on research showing that one model's outputs can be used to break the safety training of another.
Heard on the show
“The title is "LLM-Based Persuasion Enables Guardrail Override in Frontier LLMs," from a group at Maritaca AI and JusBrasil.”Episode 045 — When a Frontier Model Talks Its Own Twin Into Climate Denial