On June 10, 2026, OpenAI published its latest threat report and disclosed that it had banned two covert, China-linked influence operations that used ChatGPT to mass-produce social-media posts, comments, and AI-generated political cartoons. One cluster, which OpenAI dubbed Data Center Bandwagon, pushed the claim that the US AI data-center boom is driving up electricity prices for ordinary families. The second, Tech and Tariffs, generated cartoons attacking US trade and tech policy — with prompts that, per OpenAI's analysis, were instructed to depict President Trump while excluding China's leadership (OpenAI; CyberScoop).
The operators worked from inside China, routing around ChatGPT's unavailability there via VPNs, prompting the model in Simplified Chinese, and posing as Americans on X and YouTube. The most important number in the report is the smallest one: OpenAI rated both campaigns at the bottom of its Breakout Scale — Category 1, meaning the content never spread beyond the operation's own inauthentic accounts (the-decoder).
The case for hard regulation
Start with the strongest argument for a statutory mandate. State-linked influence operations are a textbook collective-action problem: a single platform that polices its own service still leaves the content free to migrate elsewhere, and voluntary enforcement can evaporate the moment it becomes commercially inconvenient. That is precisely the logic behind the EU's Digital Services Act (Regulation 2022/2065), which requires very large online platforms — those with more than 45 million monthly EU users — to identify, assess, and mitigate "systemic risks," disinformation among them, and to submit to independent audits (European Commission). The DSA's premise is that trust-and-safety budgets are too important to leave to quarterly priorities. Canada's newly tabled Bill C-34, introduced the same day as OpenAI's report, extends this model to AI chatbots and social platforms, with penalties reaching the greater of about US$7.2 million or 3% of global revenue (MediaNama). These are serious, good-faith attempts to make safety non-optional.
What actually stopped these operations
And yet the June 10 disclosure is awkward for the maximalist regulatory case — because no regulator stopped these campaigns. OpenAI did, on its own initiative, and it caught them at Category 1, before they reached a meaningful audience. The detection signal was not a content takedown order; it was platform-level telemetry on coordinated inauthentic behavior: account clustering, VPN patterns, and the tell-tale Simplified-Chinese prompts behind ostensibly American posts.
This matters because it locates the enforcement leverage at the right layer. The DSA and Bill C-34 regulate distribution — what platforms must remove or down-rank after content is live. OpenAI intervened at production, at the model that generated the cartoons and comments in the first place. That is a structurally earlier and cheaper choke point, and it is one no content-moderation statute currently mandates. The most effective intervention in this episode came from a firm acting on commercial and reputational incentives, not a compliance deadline.
The proportionality problem
The instinct after any disclosure like this is to legislate the response — to convert OpenAI's voluntary takedown into a legal duty for every AI provider. That instinct should be resisted in its broadest forms. The campaigns failed not for want of a statute but because abuse-detection is now a baseline expectation of operating a frontier model, enforced by competitive pressure and the threat of exactly the kind of damaging headline OpenAI got ahead of by self-reporting.
There is also a free-expression hazard in over-correcting. As the EFF warned on June 9, a growing wave of US state bills frame sweeping restrictions as child-safety or safety measures while functioning as censorship (EFF). Mandates written to stop foreign cartoons can just as easily be turned against domestic speech that a regulator finds inconvenient — and the lower the proven harm (Category 1, in this case), the weaker the justification for blunt, speech-restricting rules. Notably, the underlying messages here — that data-center demand strains the grid, that tariffs carry costs — are contested but legitimate policy arguments. The problem was the inauthentic, state-directed amplification, not the ideas. Good enforcement targets the deception, not the viewpoint.
A complement, not a substitute
The sensible reading is that platform self-policing and statutory regimes are complements working at different layers, and the regulatory layer should stay narrow and behavior-focused. Transparency mandates are the high-value, low-risk move: requiring providers to publish threat reports like OpenAI's, with consistent metrics such as the Breakout Scale, would let the public and regulators verify that self-policing is real without dictating what speech must disappear. That is closer to the DSA's audit-and-disclosure spine than to its content-mitigation duties — and far from Bill C-34's reach into chatbots.
The lesson of Data Center Bandwagon is not that regulation is unnecessary. It is that the system already worked: a private actor detected a hostile state operation early, neutralized it at the production layer, and told the public. Lawmakers tempted to bolt content-removal mandates onto AI providers should first ask what marginal safety those mandates would have added on June 10 — and whether the answer justifies the speech risk they carry.