Official OpenAI announcement Combatting adversarial distillation, dated November 20, 2026, with a Listen to article control of 13 minutes 6 seconds
Official OpenAI post Combatting adversarial distillation on openai.com, captured 21 November 2026. Image: AppFor.ai screenshot of openai.com. Image source ↗

What OpenAI published

On 20 November 2026, OpenAI published Combatting adversarial distillation. The company says it is sharing how it detects and disrupts adversarial distillation, how it has strengthened protections for its models' protected reasoning, and what it has learned that can help strengthen defenses across the industry. OpenAI defines adversarial distillation as a path for an actor to use a frontier model's outputs without authorization to train another model with similar capabilities, then deploy it without comparable protections.

According to OpenAI, its approach combines adversarial testing, model training, classifiers, and API-level protections with investigations into coordinated attempts to extract model capabilities. The post places particular emphasis on protected reasoning, the intermediate chain-of-thought steps a model generates internally, which OpenAI says can contain information the final answer leaves out.

What OpenAI reports about GPT-6.2 and enforcement

OpenAI says it has strengthened defenses against protected-reasoning extraction in GPT-6.2 Astra and GPT-6.2 Sol. In an internal adaptive evaluation using strong automated red-teaming, the company reports that the average attack success rate fell from 54.0% against GPT-6 Astra with its original safeguards to 0.2% against GPT-6.2 Astra with its updated safeguards. A table in the same post also lists 22.9% falling to 0.9% for extracting protected reasoning from the current request. Those figures are OpenAI's internal results, not an independent test in this article.

OpenAI also writes that it has seen multiple unauthorized adversarial distillation campaigns on its platform this year, and that weekly enforcement actions have ranged from thousands to tens of thousands of accounts, with the rate increasing significantly this past summer. The company says it plans to share more soon about recent campaigns it has disrupted. This article does not add campaign names or account totals beyond that official wording.

Attributions in the post, and what this article does not treat as fact

In a section on national security implications, OpenAI writes that in February it briefed Congress on activity indicating that DeepSeek was attempting to distill U.S. frontier models, and that in September it reported disrupting a campaign to extract its models' protected reasoning and attributed a core part of that activity to individuals associated with Moonshot AI, the developer of Kimi. Those statements are OpenAI's attributions. This article does not present them as independently verified findings, and it does not report any response from DeepSeek or Moonshot, because the OpenAI post does not include one.

OpenAI says it supports U.S. legislation that would give the government more tools to identify and sanction foreign actors that illicitly extract capabilities from American AI models, and that it has told Congress it supports the Chip Security Act, the AI OVERWATCH Act, and the MATCH Act. It also says it shares findings through the Frontier Model Forum and government channels, while keeping details attackers could exploit private.

Sources & further reading

Based on the official documentation below. Features and access can change; check the provider before making a decision.

  1. OpenAI: Combatting adversarial distillation, dated 20 November 2026, official page text retrieved and page rendered 21 November 2026 (opens in a new tab)
Something out of date or incorrect?Report a correction

Was this page useful?

Help us make our coverage more useful. This is feedback on the page, not a rating of the tool.

Share this page

Related reading