Update to GPT-5 System Card: GPT-5.2

OpenAI has published an update to its GPT-5 System Card, this time covering GPT-5.2. If you were expecting a dramatic overhaul of how the company handles safety for its flagship model family, you won't find one here. The document is explicit that the mitigation approach for GPT-5.2 is largely a continuation of what was laid out for GPT-5 and GPT-5.1, not a rewrite.

That continuity is the real story. OpenAI has now shipped three major iterations under the GPT-5 banner without fundamentally changing its safety playbook, which suggests the company believes its existing testing and mitigation pipeline is holding up as models get more capable. Training data sourcing stays consistent too: public web content, licensed third-party data, and material contributed by users, trainers, and researchers. Nothing new announced there, and no indication OpenAI is shifting toward radically different data practices for this release.

What's notable is what's absent from the announcement. There's no headline capability claim, no benchmark chart, no splashy comparison to GPT-5.1 or to rivals like Anthropic's Claude or Google's Gemini. This reads less like a product launch and more like paperwork, the kind of incremental documentation OpenAI is now expected to produce every time it pushes a model forward, whether the public notices or not.

That expectation itself is worth sitting with. A few years ago, a company updating an internal safety document for a minor version bump wouldn't have registered as news. Now it does, because system cards have become the closest thing the industry has to a paper trail for how these models are being tested and constrained before they reach millions of users. GPT-5.2's card being this quiet is arguably a sign the process is becoming routine, for better or worse.