Anthropic released its Mytus-class AI model, Claude Feibel5, to the general public on June 10, 2026 [1], [2].
The launch marks a significant step in the deployment of high-performance artificial intelligence, balancing raw capability with rigorous safety protocols to prevent digital misuse.
Following an announcement on June 9, 2026 [1], the U.S. startup made the model available to users globally [1], [2]. The company said Claude Feibel5 was designed to handle complex tasks while maintaining a defensive posture against potential threats [1], [2].
A central feature of the release is a set of safety safeguards intended to block malicious requests [1], [2], [3]. Rather than simply refusing a prompt, the system is designed to route requests identified as malicious to a lower-performance model [1], [2]. This approach aims to neutralize the utility of the AI for those attempting to orchestrate cyber-attacks, or other harmful activities [1], [2].
Anthropic said the goal of this architecture is to provide high-performance AI to the general public while ensuring the tool cannot be weaponized [1], [2]. By segregating high-tier processing from suspicious queries, the company seeks to maintain the integrity of the Mytus-class capabilities [1], [2].
The rollout occurs as the industry continues to debate the tension between open access to powerful models and the necessity of restrictive guardrails. The use of a tiered response system represents a specific technical strategy to mitigate risk without completely disabling the user experience [1], [2].
“Anthropic released its Mytus-class AI model, Claude Feibel5, to the general public.”
Anthropic's decision to route malicious prompts to a degraded model rather than issuing a standard refusal suggests a shift in AI safety strategy. By providing a lower-quality output to bad actors, the company may be attempting to obscure the exact boundaries of its safety filters, making it more difficult for attackers to 'jailbreak' the system through trial and error.
