OpenAI links Moonshot AI individuals to 15,000-user distillation campaign

By Splitfeed EditorialLast updated Oct 1, 2026, 10:50 UTC

On September 30, 2026, OpenAI published a report stating it had disrupted a coordinated adversarial distillation campaign involving more than 15,000 users. The company linked individuals associated with Moonshot AI, a Chinese AI startup, to the operation. According to CyberScoop, attackers used a novel encryption bypass to carry out the campaign, which targeted the internal reasoning processes of OpenAI's models. The goal, as OpenAI described it, was to extract that protected reasoning and use it to train a competing model. OpenAI said it shut the operation down and is now strengthening defenses against this class of attack. Moonshot AI has not issued any public response in the available sources.

The accusation matters because adversarial distillation, if successful at this scale, could allow a competitor to replicate the most valuable and expensive part of a frontier model without the underlying research investment. OpenAI's decision to name Moonshot AI publicly, rather than handle the matter quietly, signals that it views this as more than a routine security incident. The genuinely disputed question is whether this constitutes theft in any meaningful legal or moral sense, given that OpenAI's own models were built on data the company did not have explicit permission to use.

The Anti-AI camp is divided on emphasis. Some take the national security framing seriously: the account legal.reuters.com warned that stolen model weights could erase the US edge over China in the AI race. But several others find the complaint difficult to accept at face value. The accounts energycheckup.bsky.social, doubting-thomas.bsky.social, and dominicervolina.com each made versions of the same point, that OpenAI built its models on copyrighted material without consent and is now invoking intellectual property concerns only when its own outputs are at stake. The Pro-AI camp has not yet reacted publicly; a camp with that orientation would typically defend OpenAI's right to protect proprietary model outputs and frame the campaign as industrial espionage. The Middle Ground account techimo.bsky.social stayed close to the facts, noting that OpenAI shut the campaign down citing both security risks and unfair competition.

The argument will sharpen once Moonshot AI responds, or if regulators or courts are asked to weigh in on whether distillation of model outputs crosses a legal line that OpenAI's own data practices did not.