Anthropic CEO repeats what Ballmer said 25 years ago when calling Linux “a cancer”
In June 2001, Steve Ballmer, who had barely been CEO of Microsoft for a year and a half, granted an interview to the Chicago Sun-Times newspaper. During the course of it, he would make a historic statement by saying that “Linux is a cancer”. The curious thing is that 25 years later the CEO of Anthropic, Dario Amodei, made very similar statements when talking about how “Open Source AI is becoming a danger.” Both then and now, the reason that provoked these statements was none other than the fear that the Open Source philosophy would end up triumphing in the world. And if history teaches us anything—and perhaps Amodei should have foreseen it—it is that precisely what Ballmer did was not weaken Linux, but rather make it stronger than ever. That may also be what Dario Amodei ends up achieving. Amodei’s statements They were actually produced three years ago.. He made them in a speech before the US Senate Judiciary Committee in July 2023, but at that time they went somewhat unnoticed because at that time the most advanced AI models in the world were still very limited, and the situation for open models like Llama 3 was even worse. Linux was not dangerous per se. AI models of open weights, neither Three years later things have changed radically. The open models that several Chinese startups and technology companies have managed to develop are already very close to the impressive frontier models of Anthropic, OpenAI or Google, and Amodei’s prediction now becomes much more relevant. But it does so at a time when its Mythos and Fable 5 models have had a lot of problems precisely for being “dangerous.” Mythos Preview first and Mythos 5 now are only available for a small group of entities and companies due to its potential to find cybersecurity vulnerabilities. AND Fables 5which was a “layered” version of Mythos ended up being vetoed by the US government three days after going on the market. only yesterday its deployment was reinstatedbut it has done so with more restrictions to use it: if the model detects any dangerous intention, it is deactivated so that the user switches to using Opus 4.8. And while the US tries to put doors in the field with the excuse of national security, China does not even bat an eyelid. Chinese companies have not stopped launching more and better models of open weights, and We have the last and most splendid example in GLM-5.2the Zhipu.ai (Z.ai) model that is surprising everyone and everyone. Its creators already warned when launching it that its performance in various benchmarks is at the level of Claude Opus 5.5 or GPT-5.5. But independent analyzes in the field of cybersecurity they claim that GLM-5.2 is “as dangerous” as Opus 4.8 also in terms of cybersecurity. This points to a disturbing future for the US: that China will have models as powerful as Mythos in the short term. Jie Tang, CEO of Z.ai, agreed with that perspective: Elon Musk anticipated that these models would arrive in the first quarter of 2027, and Tang indicated that “it won’t take that long.” The real problem is not that Chinese companies develop open weight models with capabilities similar to those of Mythos. That will inevitably come, but as happened with Linux and Open Source software, The danger is that these models displace commercial software and threaten the dominant position of Anthropic and OpenAI. That’s what Ballmer feared 25 years ago, but what he seemed to point to with that FUD statement never happened. What happened was precisely what he would never have imagined: that Microsoft ended up “appropriating” Linux and Open Source solutions by integrating them into its cloud infrastructure, Azure. Today Linux virtual machines represent 61.8% of all those in Azure: this operating system has become an even more important option than Windows on that platform. It is no coincidence: the presence of Linux and Open Source platforms in the server market is absolutely dominant (about 90% globally), and the adoption of these solutions by Microsoft has been total. Not only in server environments, be careful: the Windows Subsystem for Linux (WSL) layer of Windows operating systems has been a crucial attraction for users and especially developers for years. The company made its definitive move in that section when he bought GitHub in 2018and he hasn’t looked back since. The analogy with the Anthropic (or OpenAI) situation is inevitable. Linux then threatened Microsoft’s position, and open AI models threaten that of Anthropic or OpenAI. The question here is not whether those AI models developed by Chinese companies can be dangerous: Mythos, Fable 5 and GPT-5.5/5.6 have already shown that they can be. The question is who they are for. For the world… or for the companies trying to become the de facto monopolies of this industry? Linux, after all, wasn’t a cancer. Ballmer was not right. It’s very likely that Amodei doesn’t have it either. Image | World Bank Photo Collection | Wikimedia Commons In Xataka | For decades, Linux has earned a reputation as a “shielded” operating system. Until now
