Anthropic is known for dark prophecies: the company used the completion of the Large Language Model (LLM) “myth” to warn that its capabilities were so extensive that it could not be made available to the general public. Now the company wants to have appropriate guardrails installed – with Fable 5, a secured version is released for the general (paying) public. If you want to use the current Anthropic LLM, you have to pay according to token consumption – there is no flat-rate subscription. Compared to the previous top model Opus 4.8, Anthropic doubles the prices. The announcement is accompanied by a lot of benchmarks and user reports. Anthropic quotes the FinTech company Stripe as saying that Fable 5 has compressed software development, which usually takes months, into a few days.
Myth in three degrees
The AI company only grants unrestricted access to Mythos 5 to a small number of US authorities and cybersecurity companies. In cooperation with the government, decisions are made regularly about access rights. Another variant with minor restrictions will be made available to trustworthy researchers in the fields of biology and chemistry. The public-use version of Fable 5 is available for two weeks to subscribers of the Pro, Max, Team and Enterprise flat-rate subscriptions, after which you will be billed based on usage. At a later date, Fable 5 will also be available as a flat-rate subscription.
“When AI builds itself”
In another blog post, Anthropic looks at the rapid development of LLMs and asks whether a globally orchestrated AI development freeze is appropriate. The detailed article traces the progress of the industry in recent years. Chatbots (2023) were supplemented by agents (2025), which in turn independently entrusted workers (2026) with subtasks. The fear expressed is that humans could soon be replaced at the beginning of the order chain and a self-learning and decision-making AI system would emerge.
Reminder peppered with self-promotion
This step could prove dangerous for humanity, warns the article, whose arguments repeatedly highlight the achievements of the in-house LLMs. 80 percent of the internally developed programming code is now generated by AI agents; The average developer uses Mythos to produce eight times as much code per quarter as in the years before 2025. At the same time, the report adds, many no longer have an overview of their own product, and social interaction between programmers comes to a standstill.
Demand for a brake on innovation
Due to the rapid development, Anthropic’s conclusion is that the world’s leading LLM developers agree on a general halt to further development “to give us more time to deal with the immense impact.” In the next few months, they want to confer with politicians, researchers, social institutions and other AI companies on how to deal with the current state of development.
