{"entities":[{"id":"openai","name":"OpenAI"},{"id":"openai-chatgpt-agent-system-card-2025","name":"ChatGPT Agent System Card","subtype":"Technical report","description":"As dated July 17, 2025 on its cover. Date printed on the card's cover; the dataset already holds the source record source-openai-chatgpt-agent-system-card for this PDF but no document entity.","sourceIds":["source-openai-chatgpt-agent-system-card"]},{"id":"bio-misuse-capability-evaluation","name":"Biological misuse capability evaluation"}],"claims":[{"id":"ab-ev-chatgpt-agent-bio-evaluation-categories","statement":"The ChatGPT Agent System Card states that OpenAI conducted human red teaming on biological risk as one of two key risk areas, and sets out an overview table of its biological and chemical evaluations.","derivation":"explicit","resolution":"supported","editorialState":"seed","evidenceIds":["evidence-ab-ev-chatgpt-agent-bio-evaluation-categories-1"]},{"id":"ab-ev-chatgpt-agent-card-published","statement":"OpenAI published the ChatGPT Agent System Card, dated 17 July 2025 on its cover.","derivation":"explicit","resolution":"supported","editorialState":"seed","evidenceIds":["evidence-ab-ev-chatgpt-agent-card-published-1"]},{"id":"ab-ev-chatgpt-agent-expert-deep-dives","statement":"The ChatGPT Agent System Card states that OpenAI invited seven experts with biosafety and biosecurity backgrounds to spend three to five hours with a helpful-only version of the model and to score it in a survey, and that experts noted an average success rate of about 7.7 on a scale of 1 to 10 on the tasks they attempted.","derivation":"explicit","resolution":"supported","editorialState":"seed","evidenceIds":["evidence-ab-ev-chatgpt-agent-expert-deep-dives-1","evidence-ab-ev-chatgpt-agent-expert-deep-dives-2"],"attributes":{"evaluatedSystem":"A helpful-only version of ChatGPT agent","evaluation":{"what":"How well the model performed on the tasks the experts attempted with it, and the experts' assessment of it; the card does not state who set the tasks","how":"Expert deep dives: seven invited experts used the model for three to five hours and completed a scoring survey","result":"The card states an average success rate of about 7.7 on a 1 to 10 scale on attempted tasks, with stronger performance on non-agentic than on agentic tasks","caveats":"The card states that performance on agentic tasks was weaker, which it attributes to reliability and operational issues, and that the experts' findings indicated inexperienced users would still need practical laboratory skills and critical judgement."}}},{"id":"ab-ev-chatgpt-agent-high-capability","statement":"The ChatGPT Agent System Card states that OpenAI decided to treat the launch as High capability in the Biological and Chemical domain under its Preparedness Framework, activating the associated safeguards, while stating that it does not have definitive evidence that the model could meaningfully help a novice to create severe biological harm and has taken a precautionary approach.","derivation":"explicit","resolution":"supported","editorialState":"seed","evidenceIds":["evidence-ab-ev-chatgpt-agent-high-capability-1","evidence-ab-ev-chatgpt-agent-high-capability-2","evidence-ab-ev-chatgpt-agent-high-capability-3"],"attributes":{"evaluatedSystem":"ChatGPT agent, as released on 17 July 2025","evaluation":{"what":"Whether the model reaches the High capability threshold in the Biological and Chemical domain of OpenAI's Preparedness Framework","how":"A capabilities assessment across the card's biological and chemical evaluations, reviewed under the framework by OpenAI's Safety Advisory Group","result":"Treated as High capability in the Biological and Chemical domain, with the associated safeguards activated, as a precautionary approach; the card states the decision is informed by an incremental improvement relative to previous models","caveats":"The card states there is no definitive evidence that the model could meaningfully help a novice to create severe biological harm, which is its defined High capability threshold, and that its evaluations represent a lower bound for potential capabilities."}}}],"sources":[{"id":"source-openai-chatgpt-agent-system-card","title":"ChatGPT Agent System Card","authors":["OpenAI"],"publisher":"OpenAI","publicationDate":"2025-07-17","version":"System card dated 17 July 2025 on its cover. The quoted sentence and its page match the PDF served on 2026-10-04 and the Internet Archive capture of 2025-07-17; the two copies differ only in chart labels.","evidenceUrl":"https://cdn.openai.com/pdf/839e66fc-602c-48bf-81d3-b21eacc3459d/chatgpt_agent_system_card.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20250717225133/https://cdn.openai.com/pdf/839e66fc-602c-48bf-81d3-b21eacc3459d/chatgpt_agent_system_card.pdf","archivedAt":"2025-07-17","accessedAt":"2026-10-04"}],"evidence":[{"id":"evidence-ab-ev-chatgpt-agent-bio-evaluation-categories-1","sourceRole":"primary","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-chatgpt-agent-card-published-1","sourceRole":"primary","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-chatgpt-agent-expert-deep-dives-1","sourceRole":"primary","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-chatgpt-agent-expert-deep-dives-2","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-chatgpt-agent-high-capability-1","sourceRole":"primary","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-chatgpt-agent-high-capability-2","support":"context","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-chatgpt-agent-high-capability-3","support":"context","verification":"machine_checked","retrievalRoute":"direct"}],"at":{"entities":[0,1244,1266],"claims":[3386,3387,3388,3389],"sources":[190],"evidence":[4651,4652,4653,4654,4655,4656,4657]},"version":"1lg47t7"}