Safety

OpenAI says model 'distillation' is a technical problem manageable with hardware and oversight

OpenAI and Anthropic urged US authorities to confront Chinese firms that distill their top models, a concern that prompted threats of sanctions from Trump administration officials.

OpenAI says model 'distillation' is a technical problem manageable with hardware and oversight

OpenAI and Anthropic have urged the U.S. government to act against Chinese firms that are distilling their most advanced models, framing the practice as a national security concern. Trump administration officials responded by threatening sanctions against the Chinese companies involved.

OpenAI President Greg Brockman told reporters on Thursday that distillation is not only a political or legal issue but “ultimately, there is a technical question” that can be addressed with engineering measures. He said AI will be an important priority for national competitiveness, so public concern is understandable, but solutions exist on the technical side.

What technical measures did OpenAI describe?

Brockman said OpenAI already has systems intended to prevent distillation. In a letter to Congress in February, the company outlined measures to detect likely distillation attempts. These measures include:

  • using machine learning together with human review to spot when users are scoring or evaluating model responses;
  • identifying patterns that indicate attempts to extract chains of reasoning from the models;
  • detecting large-scale generation of synthetic data, which can be a red flag for distillation efforts.

Brockman reiterated those detection and review methods during the briefing.

Price competition and a hardware strategy

Speaking around the GPT-Live demonstration and new voice capabilities, Brockman pushed back on the notion that Chinese models would undercut OpenAI on price. “It’s not the case that open source models are magically cheap,” he said. “Everything is running on the same hardware.”

OpenAI is therefore scaling its own hardware business — including chips and data centers — and positions that build-out as a competitive moat. Brockman noted that now people are paying closer attention to price, which he said was not the case three months ago, the company welcomes that because its goal is to “always be the cheapest model for any task you want.” He added OpenAI has “extremely advantageous conditions” to meet that objective.

Why this matters

Distillation touches on intellectual property, technological competition, and national security, because extracting and reusing the internal behavior of large language models can have significant implications. OpenAI’s position is that part of the response can be technical—detection, oversight and hardware controls—but political and economic measures, such as government intervention or sanctions, are also part of the broader reaction.

Brockman’s remarks were made alongside product announcements: the briefing centered on GPT-Live and new voice features, while the company also highlighted its strategic investments in hardware and data centers.