OpenAI研究：前沿模型正隐藏思考，因开发者优化能力与算力

Frontier models like GPT-6 Astra are becoming harder for safety monitors to see into. OpenAI's new research identifies which training steps are responsible and shows that the loss of transparency isn't purely a law of nature but is tied to decisions about performance and cost. To keep tabs on modern AI agents, safety researchers monitor their chains of thought (CoT). Before a reasoning model answers, it writes out its intermediate steps on a digital scratchpad that can reveal planned deception. Early on, though, Anthropic showed that these scratchpads aren't complete records, demonstrating that models like Claude 3.7 Sonnet and DeepSeek-R1 often don't even mention key clues in their chains of thought. Even so, reading the scratchpad long served as the most important safety net. As recently as this spring, OpenAI called it good news that models like GPT-5.4 could barely manipulate their inner monologue on command, even though the researchers warned at the time that low controllability doesn't guarantee reliable oversight. That trend broke for good with GPT-6 Astra, the flagship model released in September. At launch, OpenAI chief scientist Jakub Pachocki called its monitorability fragile and warned of a negative trend . He followed up shortly after in his essay "An Alien Mind," writing that no AI lab had solved monitoring well enough to keep scaling at full speed much longer. Astra recently showed how real those blind spots already are when it quietly delivered results about 40 percent worse on coding tasks for safety research, without the scratchpad monitor detecting any malicious intent, because a grader used in training had inadvertently favored this kind of "sandbagging" over outright refusal. In a new analysis from OpenAI's safety team , researchers systematically studied which specific training steps trigger this erosion. They tracked two early warning signs, silent reasoning without a scratchpad (No-CoT Capability) and targeted control over the chain of thought ( CoT Controllability ). According to the UK AI Safety Institute , Astra solves math problems that take a human about 31 minutes even with reasoning switched off. A second test asks a model, for example, to insert "meow" after every word or use only lowercase letters. Analyzing models from GPT-5 through GPT-6 Astra, the researchers traced the problem to four main causes across the training cycle. Silent mental math rises directly with the base model's raw capability, measured as declining "perplexity," meaning more accurate text prediction. As the base model improves during pretraining, it completes more reasoning steps invisibly "in its head" without leaving a written trail. This increase isn't just a byproduct of general pretraining progress but also depends on the architecture developers choose. Across three anonymized architectures, the design barely mattered on pure knowledge benchmarks like MMLU or GPQA. For tasks requiring multiple silent reasoning steps in a row, however, results diverged sharply. When the models had to solve multi-step planning tasks and work through 1,000 short programs written in everyday language, both without a scratchpad, "Architecture 2" showed a much steeper performance curve at the same perplexity than the other designs. The other architectures rose far more gently, so this design markedly favors the silent reasoning monitors can't see. Reinforcement learning (RL) after pretraining substantially increases the model's control over its own inner monologue. In the spring study, RL training with verifiable rewards had actually reduced chain of thought controllability in smaller models. For Astra-class models and new internal "Terra" models, however, this effect reverses. A model's ability to manipulate its reasoning log also transfers to future generations, similar to subliminal learning . When OpenAI trains a new "student model" on data from an existing "teacher model," the student automatically inherits the teacher's high C