Anthropic Claude.ai Enterprise via Compliance API Detection Strategy
Product Overview
Anthropic Claude App (Claude.AI) Enterprise is a tool that provides artificial intelligence (AI) productivity tooling as a service. Claude App is a popular option for enterprises leveraging the power of AI to boost productivity and efficiency for their employees in their organizations. Although organizations using AI products don’t have to worry about managing infrastructure, security is still a shared responsibility between Anthropic and their customers. Anthropic is responsible for physical security and ensuring their models and applications have necessary guardrails and built in safety checks. Organizations are responsible for ensuring they are not putting their users and data at risk with insecure configurations or compromised credentials.
The Anthropic Claude App Enterprise detection and response strategy aims to help organizations gain visibility into active threats that must be remediated. Additionally, Expel provides resilience recommendations that help organizations lock down the platform and reduce the risk of a compromise.
Expel collects data through direct Claude Compliance API integration. This integration allows Expel to collect activity fields and chat messages for detection and response.
Detection Strategy for Claude.AI
Detection
Our AI detection strategy is designed to identify and respond to suspicious user activity across AI infrastructure directly, as well as through direct integration with endpoint product AI add-on modules. By integrating directly with AI infrastructure, we continuously monitor user behavior and focus on activities such as unusual administrator configuration activity, modification of guardrails, abuse of the agent harness, or prompt abuse. These events are processed through our detection engine, which leverages behavioral analytics and threat intelligence to flag AI exploitation activity.
Anthropic does not provide native alerting for its tools. Expel will alert for the following for Claude.AI:
Destructive and/or irreversible actions by administrators against an organization’s Claude instances.
Behaviors which may indicate suspicious activity such as data exfiltration attempts, privilege escalation, malicious instructions being uploaded, or agent memory poisoning attempts. If this activity is seen coming from potentially suspicious infrastructure such as an anonymizing proxy or malicious IP address or domain, it will become an Expel alert.
Optional DUET notifications for Unsanctioned Model Context Protocol (MCP) servers, Plugins, or Claude skills.
We extend our endpoint detection strategy to detect suspected malicious activity in the agent harness. By integrating directly with EDR vendors, we can process security alerts to extract evidence and normalize event details observed during the runtime activity of an agent on a device. These normalized signals are then processed through our detection engine to look for signs of post-exploitation activity.
When a threat or anomaly is detected, our automated response bot, Ruxie, takes action by enriching evidence fields with first- and third-party threat intelligence. Additional Ruxie actions query a wide span of technologies directly to arm analysts with key pieces of investigative information and related events.
Response
Integrations with AI infrastructure and endpoints provide rich context for processes and also support other types of Expel Alerts. For example, we use source device identification across a number of alert types when a source IP or hostname is available, because it provides rich context about the actor behind the activity.
Additionally, AI infrastructure and endpoints provide valuable information for network alerts to help identify which process triggered a connection.
To learn more about our overall approach to detection strategy, see About Detection Strategy.
What We Support for Anthropic Claude.ai Enterprise
To see a comprehensive list of the most up-to-date Expel detection rules, vendor detection rules, opt-in detections, and available DUETs (did you expect this) that we support for Anthropic Claude.ai Enterprise, you can visit the Detections page in Workbench or ask your Sales or Support rep for the most recent download.
Supported event log sources
Activity field logs
Prompt messages
Anthropic Claude.ai Enterprise detection rules support
No native Anthropic detection support exists.
Detection rules written by Expel
Yes.
Investigative support through Workbench
Yes. We are able to take the following investigative actions to gather data for triage and investigation of events:
Query Logs
Query User
Query IP
Related vendor alerts
Get chat history
Hunting support
No. Hunting is not currently available for this integration.
Additional Details and Common Questions
DUET
A DUET rule flags certain events as needing an immediate verification or notification, and bypasses the normal internal event triage process. The events subject to DUET rules contain behaviors that are not typically indicative of true security incidents, as they are related to policy violations or potential risk.
There are a number of workflows that a DUET may follow. When enabled, the activity will be flagged for investigation and will be routed to you (rather than to us) to take a specified first action. To see the specific DUET rules currently supported for this integration, visit the Detections page in Workbench.
Last updated
Was this helpful?

