Constrain external content access for agents
Limit agent access to external content sources to reduce risk from untrusted data.
Control the external content an agent can reach to mitigate risks from untrusted sources.
Tools like MCP servers, third-party plugins, and web search feed content into an agent's context from sources you don't control. Even audited connectors can load poisoned data. Granularly limiting tool permissions, such as granting read-only DB access instead of write access, helps limit the blast radius.
An agent with read-only database access can be deployed more broadly than one that writes to production.
Manage agent risk by limiting access to external content. Be aware that even audited connectors can ingest malicious data. Use granular tool permissions to restrict what the agent can interact with, reducing its potential blast radius.
- How we contain Claude across products ↗Anthropic Engineering · Anthropic Engineering · Jun 5, 2026 · paraphrased
First seen on Subvolts Jun 5, 2026. Updated Sep 11, 2026. Spotted a problem? Tell us.
- How we contain Claude across productsAnthropic · Anthropic Engineering · Jun 5, 2026
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- HeuristicContain agents by limiting what they can doSupervise agent capabilities via access boundaries like sandboxes and VMs, not just behavior. · 15 min
- Anti-patternAgents acting beyond intended scope or contextAgents may take actions that seem reasonable to them but are not what humans intended due to lack of context. · 5 min
- WorkflowBuild an AI agent evaluation pipeline in one hourCompress weeks of AI agent testing into one hour using open-source tools and automated pipelines. · 60 min
- HeuristicAI agents need time-awareness for runtime controlAI agents require a sense of time to predict, estimate, and control their own actions and execution duration. · 5 min
- HeuristicComposable controls for agent autonomy and safetyUse composable controls that perform single jobs, enabling agents to run with autonomy at appropriate safety levels. · 10 min
- HeuristicDefine agent approval loops as product design decisionsFrame human-in-the-loop requirements for agents as product design choices, not technical limitations. · 10 min
Questions people ask
- Which AI tools does this work with?
- Any current model. It was seen with in the sources.
- Where did this come from?
- Anthropic Engineering (Anthropic Engineering). Subvolts collected, normalized and credited it; the link above opens the original.