Content Standard Companies
The AI 'Do Not Train' Standard: Meaningful Norm-Setting or a Toothless Gesture?
Google, OpenAI, Microsoft, and Anthropic have all backed a proposed standard from the Internet Engineering Task Force that would allow website publishers and content creators to signal that they do not want their content used to train AI models. On its face, the announcement reads as a meaningful concession to journalists, publishers, and independent creators whose work has been absorbed into AI training pipelines without compensation or consent.
The caveats, however, are extensive. Compliance is voluntary, with no enforcement mechanism — a framework that critics immediately compared to the 'Do Not Track' browser header introduced around 2009, which virtually every advertising network ignored because there was no legal obligation to honor it. More consequentially, the directive cannot undo past collection: content already scraped and used in training data remains in the companies' hands, offering nothing retroactively to publishers whose work built the models now competing with them.
Meta's posture is instructive. The company said it 'supports the idea' of the opt-out standard but has not committed to implementing it — a form of non-commitment that avoids the bad optics of outright opposition while preserving complete operational flexibility. Meta's AI training practices are currently under legal challenge in multiple jurisdictions, including a significant case in the EU, making the reluctance to commit appear strategic rather than principled.
A more charitable reading holds that the standard's value may lie not in self-execution but in legal scaffolding. By publicly endorsing the concept of content-creator opt-out rights, the four companies make it harder to argue in future litigation or legislative proceedings that the industry was unaware publishers wanted control over their content. Whether courts in cases such as The New York Times v. OpenAI begin citing the IETF proposal as evidence of an emerging industry norm — and whether that argument gains any judicial traction — may be the clearest indicator of whether the gesture does real work or simply shifts the conversation.
Stress-Testing the Consensus: What If the Critics of Voluntary AI Standards Are Wrong?
The dominant critical position on the IETF opt-out standard — shared by media lawyers, academic critics of AI, and content-creator advocacy groups — is that voluntary industry standards are inherently inadequate and the proposal is essentially performative. The confident version of that argument rests on assumptions worth examining directly.
The first assumption is that industry actors will not honor commitments without legal compulsion. The counterexample most readily available is robots.txt, a protocol roughly analogous to what the IETF is proposing. Major web crawlers including Google and Bing have voluntarily honored robots.txt for decades without any binding enforcement mechanism, because doing so maintained the trust relationships with publishers that made their search products useful. The question is whether AI training data has the same incentive dynamics — and there are reasons to think it does not. Search engines had an ongoing dependency on publisher cooperation; AI models are trained in discrete cycles, and a completed model's value does not depend on continued access to any individual publisher's content. The incentive to honor an opt-out is structurally weaker.
A more productive framing may be that voluntary standards and legislation are not alternatives but sequential steps. If the IETF proposal establishes that publishers have expressed a clear, documented preference — strengthening plaintiffs' hands in copyright litigation and legislators' hands in drafting mandatory licensing frameworks — its value lies not in self-execution but in what it enables downstream. The specific signal to watch: whether courts in AI copyright cases, most prominently The New York Times v. OpenAI, begin citing the IETF proposal as evidence of an emerging industry norm that the defendant failed to honor. If that argument appears in a legal brief and gains judicial traction by mid-2027, the voluntary standard is doing real work. If it has been cited in zero proceedings by that point and no major AI company has changed its training data practices, the cynical read was correct.
A second assumption embedded in the Google education coverage deserves equal scrutiny: that school districts will push back meaningfully against the Search restriction rather than simply capitulate. Platform leverage depends entirely on whether the leveraged party has credible alternatives. For most school districts, operationally, those alternatives do not currently exist. If districts facing the actual reality of losing Google Search quietly accept the AI terms rather than fight them, Google learns that the strategy works — and the template becomes available for application in healthcare systems, municipal government contracts, and any other sector with deep Google infrastructure integration. The indicator to track: how many districts that publicly objected to the policy are still objecting six months from now.