Version history

OpenAI updated their OpenAI GPT-5 System Card on September 30, 2026. Change detected: 2 sentence(s) modified. Document contained 13 sentences after update.

OpenAI updated their OpenAI GPT-5 System Card on September 26, 2026. Change detected: 1 sentence(s) modified. Document contained 13 sentences after update.

OpenAI updated their OpenAI GPT-5 System Card on September 24, 2026. Change detected: 1 sentence(s) modified. Document contained 13 sentences after update.

OpenAI updated their OpenAI GPT-5 System Card on September 23, 2026. Change detected: 1 sentence(s) modified. Document contained 13 sentences after update.

OpenAI updated their OpenAI GPT-5 System Card on September 17, 2026. Change detected: 1 sentence(s) modified. Document contained 13 sentences after update.

OpenAI's GPT-5 System Card was reformatted to include the document title 'GPT-5 System Card | OpenAI' at the beginning. The underlying description of GPT-5's unified system architecture, reasoning models, and routing behavior remains unchanged. This is a presentational modification with no change to the stated system capabilities or safety documentation.

OpenAI's GPT-5 System Card was revised in an update detected on September 9, 2026. The change removed the header 'GPT-5 System Card' from the opening line and updated the 'Keep reading' section, replacing three related articles (An Alien Mind, Research acceleration, and Safety overview) with three different articles dated September 6-8, 2026. The operational significance is limited; this appears to be a routine update to the document's internal cross-references and navigation elements rather than a substantive change to the model's capabilities, safety thresholds, or disclosed risks.

OpenAI removed a reference to a security incident article ('The Hugging Face incident and the road ahead') from the GPT-5 System Card on September 7, 2026, and replaced it with links to three other safety and research-related articles. The core safety assessment language regarding biological harm capability remained unchanged. This appears to be a editorial reorganization of related documentation links rather than a substantive change to safety claims or technical content.

OpenAI revised the GPT-5 System Card on September 5, 2026 by updating internal content references and links. The change removed a reference to 'Offering Zero Data Retention for frontier models' and replaced a link to 'Path to Astra: critical capabilities and frontier safeguards' with a new 'Safety overview: GPT-6 Astra Safety' resource. These modifications reflect updated internal documentation navigation and do not alter the substantive safety assessment or risk evaluation language of the System Card itself.

OpenAI's GPT-5 System Card was updated in an update detected on September 2, 2026. Two sentences were modified: the precautionary approach statement on biological harm remained substantively unchanged, and a navigation link titled 'Path to Astra: critical capabilities and frontier safeguards' dated September 1, 2026 was added, while a link about 'Pacing model development in an era of cyber-critical capabilities' dated August 18, 2026 was removed. This represents a restructuring of the document's referenced safety resources rather than a change to the core safety assessment or methodology.

OpenAI's GPT-5 System Card was revised in an update detected on August 27, 2026. The change involved reformatting the document header and updating the related articles section. Specifically, the 'Pacing model development' and 'Offering Zero Data Retention' articles remain, but a new linked article titled 'The Hugging Face incident and the road ahead' (dated August 26, 2026) was added to the related content section.

OpenAI removed the title 'GPT-5 System Card' from the header line of their system card document in an update detected on August 21, 2026. The document previously opened with 'GPT-5 System Card | OpenAI August 7, 2025' and now opens with only 'OpenAI August 7, 2025'. The substantive content describing GPT-5's architecture and routing capabilities remains unchanged.

OpenAI's GPT-5 System Card was updated in an update detected on August 20, 2026. The document removed a reference to an August 18, 2026 article on 'Introducing ChatGPT for Teens' and replaced it with references to two articles: one on 'Offering Zero Data Retention for frontier models' dated August 19, 2026, and the retained 'Pacing model development in an era of cyber-critical capabilities' article. The substantive risk assessment language regarding biological harm remained unchanged. This change appears to reflect editorial updates to referenced materials rather than modifications to the model's risk evaluation or governance stance.

OpenAI's GPT-5 System Card was updated on August 19, 2026 to replace references to three previously linked articles with two newer articles dated August 18, 2026. The change substitutes older content links (published August 1, 10, and 29, 2026) with current references to 'Pacing model development in an era of cyber-critical capabilities' and 'Introducing ChatGPT for Teens'. This is a documentation refresh reflecting current published materials rather than a substantive change to the safety assessment or capability evaluation contained in the system card itself.

OpenAI's GPT-5 System Card was updated in an update detected on August 11, 2026. The change involved replacing one featured publication link in the document's related content section. The previous link directed to '2025 System Cards', while the updated version links to 'Expanding Daybreak as the Cyber Defense Window Narrows'. This is a formatting and content curation change with no material impact on the safety assessment, capability claims, or operational provisions of the system card itself.

OpenAI revised the related-content links in their GPT-5 System Card detected on August 2, 2026. The document previously linked to a July 20, 2026 publication on 'Safety and alignment in an era of long-horizon models' and a July 28, 2026 publication on 'Scientific computing in the age of agentic AI'. The updated version replaces the July 20 publication link with a new August 1, 2026 publication on 'Ten advances in mathematics and theoretical computer science'. This change affects which research publications are highlighted as related safety and capability materials in OpenAI's risk documentation.

OpenAI updated the related-articles section of its GPT-5 System Card detected on July 30, 2026. The document previously linked to three articles on safety topics published between July 15-20, 2026. The updated version replaced those articles with three different links on AI benchmarking and scientific computing published July 28-29, 2026. This is a formatting and editorial change to the document's supplementary reading list with no material change to the system card's substantive safety assessment or capabilities analysis.

OpenAI updated the related-content section in its GPT-5 System Card detected on July 21, 2026. The previous version linked to an article titled 'OpenAI Bio Bug Bounty' from July 9, 2026 as one of three featured safety articles. The updated version replaced that article with 'Safety and alignment in an era of long-horizon models' from July 20, 2026. This changes which safety research and discussion OpenAI surfaces as contextually relevant to its biological-harm risk assessment.

OpenAI's GPT-5 System Card was updated on July 18, 2026, with a modification to the document's linked content references. The change replaced references to three previously linked articles ('Separating signal from noise in coding evaluations', 'Introducing GeneBench-Pro', and a generic 'View all' link) with two new articles titled 'Why teens deserve access to safe AI' and 'GPT-Red: Unlocking Self-Improvement for Robustness'. The substantive text of the system card remains unchanged; only the curated set of related documents and safety research linked from the card has been updated.

OpenAI updated the related-content links in the GPT-5 System Card on July 13, 2026. The previous version linked to three research articles (GeneBench-Pro, LifeSciBench, and an older 2025 System Cards reference). The updated version replaced two of those links with newer content, including an OpenAI Bio Bug Bounty Safety announcement from July 9, 2026 and a coding evaluation research piece from July 8, 2026. This change does not alter the substantive risk assessment or safety conclusions stated in the document itself.