Tag: red teaming

Published on August 8, 2024This article details the safety measures undertaken before the release of GPT-4o, including external red teaming and frontier risk evaluations. Key risk areas such as voice generation, speaker identification, and unauthorized content generation were evaluated, with appropriate mitigations implemented. GPT-4o’s voice capabilities, which closely match human response times, were assessed as…