Anthropic Ships Sonnet 5 With Agentic Safety and Pricing Details
1 July 20265 min read
AI
Anthropic Updates Its Model Spec to Address Agentic and Multi-Model Deployments
Anthropic has revised its published model specification to include guidance on how Claude should behave when operating as an agent, receiving instructions from other AI systems, and acting with greater autonomy in multi-step tasks.
Illustrative image. Cedar S. Insights uses editorial stock photography; images do not depict specific events described in articles.
Anthropic has published an updated version of its model specification, the document that describes the values, priorities and behavioural guidelines the company uses when training Claude. The revision adds substantial new guidance on agentic use cases, which have grown significantly since the original specification was published.
The updated spec addresses situations where Claude operates as an agent within a larger system, receiving instructions from orchestrating software or other AI models rather than directly from a human user. Anthropic says Claude should apply the same ethical principles regardless of whether instructions come from a human or another model, and should be appropriately sceptical of claimed permissions that were not established in the original system prompt.
The document also addresses the concept of minimal footprint: the idea that agents should request only the permissions they need, prefer reversible actions over irreversible ones, and err on the side of checking with a human when uncertain about the intended scope of a task.
Anthropic describes a hierarchy of principals — the company itself, operators who deploy Claude through the API, and end users — and explains how Claude should navigate conflicts between their instructions. The company says operator instructions generally take precedence over user requests, but neither can override Anthropic's core guidelines.
A model specification describes intended behaviour, not guaranteed behaviour. The degree to which a trained model reliably follows its specification across diverse real-world inputs is an empirical question that requires ongoing evaluation.
Why It Matters
Model specifications are becoming an important form of AI governance documentation. As AI systems take on more autonomous roles, the question of what values and constraints are built into them — and how those are communicated to deployers and users — becomes more consequential. Anthropic's decision to publish and update its spec publicly creates a degree of accountability: the company's stated intentions can be compared against observed model behaviour.
Primary Sources
Our sourcing: Cedar S. Insights provides source-led editorial analysis. Reported company, institutional and regulatory claims are attributed to their original sources unless stated otherwise.
Corrections: If a material factual error is identified, Cedar S. Insights will update the relevant article and preserve the distinction between the corrected statement and supporting evidence.
Topics
AnthropicClaudeAI SafetyModel SpecAI Agents