Anthropic Releases Claude Opus 5.5 With Prompt-Injection Safeguards
Summary
Anthropic released Claude Opus 5.5 with stronger prompt-injection defenses, a coding-action classifier and an auditable sandbox. The company says it attempted to cross containment boundaries about 85% less often than earlier models in an evaluation.
Key points
- Anthropic says the model attempted to cross containment boundaries about 85% less often than Opus 5 or Claude Mythos 5.1 in one evaluation; all attempts were rated low severity and self-reported.
- A classifier screens coding-agent actions before execution, and Anthropic introduced an open-source sandbox for security teams to audit.
- Preserved thinking is designed to prevent API users from editing prior context and make large-scale model-distillation attacks harder; it applies to qualifying API accounts created on or after August 31, 2026.
- Anthropic says most cybersecurity tasks are rerouted to Opus 4.8 and plans to broaden Opus 5.5 access for verified cybersecurity professionals.
- Anthropic cautions that current evaluations cannot identify every potential failure before deployment.
Article Details
- Event Type
- AI model release
- Impact
- Anthropic released Claude Opus 5.5 across its platforms and major cloud providers. The model adds safeguards for prompt-injection attacks, model copying, and sensitive work, while offering lower API prices and faster output. The article reports no specific security incident or confirmed compromise.
People
Vendors
Amazon Web ServicesClaude Opus 5.5 is available across Anthropic’s platforms, Amazon Web Services, Google Cloud and Microsoft Azure.AnthropicClaude Opus 5.5 is now available across Anthropic, AWS, Google Cloud, and Microsoft Azure, with watermarking and preserved-thinking safeguards designed to support EU AI Act compliance and reduce model-copying risks.Google CloudClaude Opus 5.5 is now available across Anthropic, AWS, Google Cloud, and Microsoft Azure, with watermarking and preserved-thinking safeguards designed to support EU AI Act compliance and reduce model-copying risks.Microsoft AzureClaude Opus 5.5 is now available across Anthropic, AWS, Google Cloud, and Microsoft Azure, with watermarking and preserved-thinking safeguards designed to support EU AI Act compliance and reduce model-copying risks.
Products
Claude CodeA separate fast mode is available through Claude Code and the Claude Platform, offering up to 2.5 times the speed for $8 per million input tokens and $40 per million output tokens.Claude Mythos 5.1In an evaluation of attempts to cross containment boundaries, the model tried to circumvent its assigned limits about 85% less often than Opus 5 or Claude Mythos 5.1.Claude Opus 5Claude Opus 5.5 is now available across Anthropic, AWS, Google Cloud, and Microsoft Azure, with watermarking and preserved-thinking safeguards designed to support EU AI Act compliance and reduce model-copying risks.Claude Opus 5.5Claude Opus 5.5 is now available across Anthropic, AWS, Google Cloud, and Microsoft Azure, with watermarking and preserved-thinking safeguards designed to support EU AI Act compliance and reduce model-copying risks.Opus 4.8Most cybersecurity tasks are rerouted to Opus 4.8.
Countries
Industries
BiologyHe described this approach as a way to manage risks, remain competitive with China and deliver benefits in fields such as biology and medicine.MedicineHe described this approach as a way to manage risks, remain competitive with China and deliver benefits in fields such as biology and medicine.