OpenAI accuses DeepSeek of 'free-riding' on US models

OpenAI is taking its grievances with DeepSeek straight to Capitol Hill.

According to a memo seen by Reuters, Sam Altman’s team warned the US House Select Committee on Strategic Competition between the US and the Chinese Communist Party that Chinese startup DeepSeek has been systematically copying ChatGPT and other top American models.

OpenAI accused DeepSeek of "ongoing efforts to free-ride on the capabilities developed by OpenAI and other US frontier labs," claiming the Hangzhou-based lab bypassed standard bans to train its own systems.

Bypassing limits with third-party routers

At the center of OpenAI's argument is model distillation, a process where a smaller or newer model learns directly from the outputs generated by a larger, more established model.

OpenAI told lawmakers it spotted accounts tied to DeepSeek employees actively engineering workarounds to evade platform blocks. They allegedly routed traffic through obfuscated third-party proxies to mask where the requests were coming from.

The memo didn't stop there. OpenAI claimed it knows DeepSeek employees wrote code specifically to query US models at scale, gathering massive batches of output programmatically to feed its distillation pipelines.

DeepSeek and its hedge-fund parent company, High-Flyer, did not immediately comment on the report.

Accusations of cutting corners

This isn't just about terms of service violations. OpenAI told lawmakers that Chinese labs are "actively cutting corners when it comes to safely training and deploying new models."

OpenAI says it routinely kicks users off the platform whenever it catches them using outputs to build competing tools. But playing whack-a-mole with automated scrapers behind proxy networks has proved notoriously tough.

The timing is especially sensitive given that DeepSeek rattled the global tech sector early last year. Its DeepSeek-V3 and reasoning-focused DeepSeek-R1 models matched top US benchmarks at a fraction of the reported cost, winning praise from several Silicon Valley leaders while triggering alarm bells across Washington.

Why it matters

For builders and researchers, this fight drags one of the industry's open secrets into the regulatory crosshairs. Distillation has always been a gray area in AI, but relying on rival API outputs to train direct competitors carries growing legal and access risks.

For the industry, OpenAI’s memo shows that the front lines of the AI race are shifting into politics. We think this is a deliberate push by OpenAI to enlist lawmakers where technical defenses fall short: if you can't stop automated routing across open web APIs, you push for tighter export and compliance controls instead.

At the same time, it highlights the fragility of closed-source moats. If frontier intelligence can be extracted simply by querying public endpoints, model safety and proprietary advantages become nearly impossible to defend without walling off access entirely.