Anthropic re-routes sensitive security requests to older model

Anthropic recently introduced its newest AI model, Claude Opus 5.5, which promises better reasoning and lower operational expenses. According to report details from The Decoder, the model matches the capabilities of previous top tier offerings while cutting token costs significantly and improving response speed. The release aims to address common user feedback regarding high pricing and overly complex language generation.

To handle safety concerns around potential misuse, Anthropic has implemented a specialized request handler. As reported by The Decoder, when a user asks the new model to perform sensitive cybersecurity tasks, the system automatically redirects the query to an older version called Opus 4.8. Similarly, inquiries related to advanced biological research or automated software development are handed off to Opus 5 unless the user belongs to a vetted organization with approved access.

This setup allows Anthropic to offer advanced engineering tools without giving unrestricted access to features that could be weaponized. While safety researchers often praise proactive restrictions, routing users to older models without clear warnings could frustrate software developers who legitimately need modern tools to patch digital vulnerabilities. It remains to be seen whether this redirection system will become standard practice across competing AI vendors.