OpenAI takes the distillation fight to Washington

On 12 February 2026 OpenAI warned US lawmakers that DeepSeek was using sophisticated methods to extract results from leading American AI models in order to train the next generation of its R1 system. It was the moment the distillation argument stopped being an industry grievance and became a lobbying position.

The specific claim

OpenAI characterised the activity as part of ongoing efforts to “free-ride on the capabilities developed by OpenAI and other US frontier labs.” More consequentially, it said it had detected new, obfuscated methods designed specifically to evade the defenses it had put in place after the original 2025 DeepSeek dispute.

That detail is the important one. It implies an adversarial cycle rather than a one-off incident: a defense is deployed, it is characterised, it is worked around. Every subsequent development in 2026 followed that pattern.

Why February mattered

The 2025 accusations against DeepSeek had been argued largely in public and in the press. Taking the claim to legislators changed the venue and the stakes. It framed distillation as a matter for policy rather than for terms of service, and it set up the sequence that ran through the rest of the year: a second lab joining the complaint within a fortnight, formal correspondence to a Senate committee in June, and White House statements in July.

What was not provided

OpenAI did not publish the evidence underlying the claim, and the “obfuscated methods” were not described in technical detail. The pattern of asserting detection without publishing the detection method recurred throughout the year, and is the reason the dispute has remained argumentative rather than evidentiary.