It’s just Anthropic’s marketing.

Claude Mythos is making waves. Initially, the preview was only made available to a very select few Big Tech companies, such as Google and Apple. It seems Anthropic decided it would be too dangerous to release it to the public all at once. The access was gradually expanded, and recently, news broke in my country, Japan, that a major megabank had been granted access. Politicians are also treating it as a risk, and it feels like discussions about how to use AI safely have been increasing lately.

Even among the general public, especially on X, there are voices expecting that Claude Mythos will change the world or that it might achieve AGI. However, they are missing a crucial perspective: "This includes Anthropic’s marketing."

When AI companies release a model, they often use the tactic of developing it themselves and then declaring, "It's too dangerous," before rolling it out in stages. This was exactly the case with GPT-3.5 and GPT-5. I was particularly worried that GPT-5 would fundamentally upend the way the world works, and I remember staying up until 3:00 AM (Japan time) watching OpenAI’s announcement with heavy eyes.

My reaction after watching it? "What is this? It’s not much different from what we already have." My initial impression was wrong. When I actually tried it, the GPT-5 of that time actually felt less capable overall than GPT-4o.

Before GPT-5 was released to the world, it was already being used internally at OpenAI. It was said that GPT-5 was handling most of the internal operations, and that "once this is released, it’s going to be amazing!"

Certainly, there are impressive models in the GPT-5 series. I currently use version 5.5 for AI agents. It is far more efficient than previous models, but it didn't result in a situation that fundamentally overturned the structure of society. Today, I am still here writing code, writing tests, and submitting pull requests.

Getting back to the topic of Mythos—this time it’s about Claude, or rather, Anthropic. Until now, Anthropic often released features and new models smoothly without resorting to this kind of hype. But this time, it seems they’ve resorted to hype-based marketing. "We found a vulnerability that no human has found in 27 years!" Seriously? Please tell us how significant the impact of this vulnerability actually was.

In Japan, it is said that the systems connecting most financial institutions' networks are still written in COBOL. It would be unnatural if they didn’t have vulnerabilities. (I do welcome it if this actually helps those systems.)

Anthropic is preparing for an IPO. Their valuation changes in an instant based on their hype. There is no reason for us, as general developers or business people, to be swayed by Anthropic’s marketing.

Going slightly off-topic, I’m annoyed by Anthropic’s "entitled" business practices. For example, not allowing users to use their regular subscription plans with third-party tools like OpenCode. Or how they took Bun—a tool developed by a subsidiary—and completely changed the language via something close to vibe-coding without a single review in less than a week.