The founding dream of the digital revolution was that of a digital commons. It sounds romantic after we have watched the commons being sliced, packaged, sold, and commodified. And yet, it is a dream I want to hold on to.There were extraordinary visions of what digital technologies could do if they were built as public infrastructures. Public does not mean free or outside power. The QR code was patented but made publicly usable; Linux, open protocols, and community-maintained code still hold up much of the digital economy. The giant private economy of the internet was powered by infrastructures it did not fully own.AdvertisementFor a long time, however compromised, the digital commons rested on three sides: The people who built and maintained its infrastructures; the publics who used, adapted, contested, and animated its applications; and governments, which were supposed to regulate, safeguard, and hold these systems accountable to collective progress.Large language models have destabilised all three. These systems are built on public infrastructures, public data, public labour, open technical cultures, and decades of shared digital experimentation. Their governance, however, is black-boxed and negotiated behind closed doors. Their financial logic is no less alarming. Anthropic has announced a $380 billion post-money valuation. OpenAI has been reported at a valuation of $852 billion. These valuations do not emerge from settled public trust, social safety, or proven sustainability. They emerge from the conversion of collective dependency into private speculative power.With Claude Mythos, we are perhaps not only seeing the end of the dream, but its replacement.AdvertisementAnthropic has made Claude Mythos Preview available through Project Glasswing as a gated research preview, ostensibly to help selected partners find and fix vulnerabilities across critical software systems. Anthropic also says Mythos Preview has already identified thousands of zero-day vulnerabilities. On paper, this sounds like responsibility. In practice, it shows us the problem.A private corporation, built on public infrastructures, is now deciding who gets access to a model powerful enough to find vulnerabilities in the systems on which the world depends.The question of whether AI regulation should be heavy-handed or with a light-touch has arrived too late. Claude Mythos signals the end of the internet as a public utility, and the shrinking power that governments have in national and global regulation.Who watches the watchman? Nobody. With Mythos, the watchman who is supposed to identify the vulnerabilities and cracks is also the only one who can protect the system by taking an ethical position. Except Mythos has no sense of ethics. Anthropic might loosely have one, but not enough to stop it from releasing the model, even in limited preview.Technologies are not tools. They are propositions. They imagine the worlds that cluster around them, and then begin to make those worlds available, profitable, and inevitable. They are not dangerous only when deployed. They can be dangerous in their very existence. We now have the nuclear version of generative AI sitting with us. Not because it will destroy the world tomorrow, but because its existence reorganises the conditions of security, vulnerability, trust, and power. Anthropic might stall this for a while, but others will get there and not perform this limited restraint. Once a capacity exists, it becomes normalised, and we adapt to it.Regulation and oversight are of little use if they are always playing a chasing game. AI companies have turned regulation into the last line of redress when it should have been the first line of defence.you may likeAsking whether we need more or less regulation to keep people safe is like asking whether regulation can deweaponise a gun. In a world where the gun exists, people will die. Regulation may reduce the damage, but it cannot undo the world that the weapon has already made.What remains is to ask whether global governance systems can do what regulation can no longer do: Not merely control these technologies, but slow them, refuse them, perhaps force their degrowth. Control might now be a fantasy, as naive as the dream of the digital commons. But there is still a difference between fantasies. The commons asked us to build public power, however imperfectly. This new fantasy asks us to trust the people who broke the world, and then handed us repair as a service.The writer is professor of Global Media at the Chinese University of Hong Kong and faculty associate at the Berkman Klein Centre for Internet and Society, Harvard University