Claude's owner launches the company's most powerful AI to the public, but imposes restrictions

Anthropic
Reuters via BBC
Anthropic, the American company responsible for the Claude chatbot, announced this Tuesday (9) the launch of Claude Fable 5, described by the company as the most powerful artificial intelligence model ever made available to general users.
The system is part of the Mythos class, a new generation of models presented by the company in April and considered superior to the Opus family, which until then represented the company's most advanced level.
According to Anthropic, Fable 5 achieves cutting-edge performance in areas such as software engineering, data analysis, scientific research, computer vision and complex reasoning tasks.
Instagram Plus is released in Brazil; see price and benefits
Despite opening to the public, the company decided to impose restrictions on topics considered sensitive.
When users make requests related to cybersecurity, biology, chemistry or knowledge extraction techniques from AI models, the system can automatically transfer the conversation to a less powerful version, called Claude Opus 4.8.
According to Anthropic, the measure was adopted because Mythos class models have reached a level of capacity that could pose significant risks if used maliciously. The company claims that these technologies are particularly effective in discovering and exploiting software vulnerabilities, which could facilitate more sophisticated cyberattacks.
The company claims that more than 95% of usage sessions should not be affected by the restrictions, but recognizes that some legitimate requests may end up being blocked out of an abundance of caution.
Unrestricted version
At the same time, Anthropic released Claude Mythos 5, a version of the same model with part of the safeguards removed.
Initially, access will be restricted to a select group of partners linked to cyber defense and the protection of critical infrastructure through the Project Glasswing program, developed in collaboration with the United States government.
The company said it intends to gradually expand access through a trusted user program.
According to Anthropic, Mythos 5 has the most advanced cybersecurity capabilities among currently available artificial intelligence models and can identify and exploit software flaws with unprecedented speed and accuracy.
Applications in science
The company also highlighted advances in the model in scientific research. According to Anthropic, Mythos 5 was able to accelerate stages of drug development, assist in the design of proteins and generate unprecedented hypotheses in molecular biology that began to be experimentally evaluated by researchers.
The company also states that the technology was able to conduct research in genomics in a largely autonomous way for more than a week, analyzing millions of cells from different animal species and developing its own machine learning models to interpret the data.

Source: G1