Abnormal AI Agent Security
Product · SaaS / cloud software · HR & recruiting by Abnormal AI
Keep Every AI Agent on Task
Abnormal AI Agent Security provides governance and protection for autonomous AI agents by mapping them to owners and behavioral baselines. It enables security teams to inventory agents across major AI platforms and automatically contain rogue behavior.
Category: Cybersecurity
- Deployment
- Cloud-based API integration
- Platforms
- Vertex AI, Amazon Bedrock, Microsoft Copilot
- Integrations
- Workday, Salesforce, ServiceNow
- Supported Platforms
- Vertex AI, Amazon Bedrock, Microsoft Copilot
- SaaS Integrations
- Workday, Salesforce, ServiceNow
- Core Engine
- Attune AI
- Inventory of agents across Vertex AI, Bedrock, and Copilot
- Maps every agent to its owner and access permissions
- Detects deviations from behavioral baselines
- Automated suspension of rogue agent runs
- Revocation of overprivileged credentials
Overview
Abnormal AI Agent Security is a governance and protection solution designed for enterprises deploying autonomous AI agents. It addresses the risk of agents acting beyond their assigned tasks by connecting every agent to its owner, access permissions, and a behavioral baseline. The platform maintains a live inventory of agents across platforms like Vertex AI, Amazon Bedrock, and Microsoft Copilot, mapping their credentials to the sensitive resources they can reach. By detecting deviations from expected behavior, such as a research agent querying payroll data, it allows security teams to automatically suspend rogue runs or revoke credentials. This API-based solution integrates with SaaS platforms like Salesforce and Workday to monitor agent interactions and ensure they remain on task.
Governance and Visibility
- Agent Inventory: Maintains a live inventory of connected AI agents and the platforms they run on.
- Ownership Mapping: Connects every agent to its owner, access permissions, and normal behavioral patterns.
- Access Mapping: Visualizes the credentials, tools, permissions, and sensitive resources each agent can reach across the organization.
- Behavioral Baselining: Establishes a behavioral baseline for each agent to identify deviations that put data or systems at risk.
Protection and Containment
- Run Suspension: Suspends agent runs or ends sessions based on fine-grained response policies.
- Credential Revocation: Revokes agent credentials automatically when rogue activity is detected.
- Containment Verification: Verifies containment and allows security teams to review agent actions taken before the stop.
- Anomaly Detection: Flags activity outside an agent's normal behavior, such as a research agent querying payroll data, for investigation.
Platform Connectivity
- AI Platform Support: Monitors agents deployed on AI platforms including Vertex AI, Amazon Bedrock, and Microsoft Copilot.
- SaaS Integration: Connects with SaaS platforms like Workday, Salesforce, and ServiceNow to monitor agent interactions.
- Attune AI Integration: Uses the Attune AI engine to learn normal behavior for non-human identities and service accounts.
Use cases
- Security teams inventory all AI agents across Vertex AI and Bedrock to identify owners and access paths.
- Administrators detect when a research agent begins exporting sensitive payroll data outside its normal behavioral baseline.
- Security operations automatically suspend a rogue agent run and revoke its credentials based on fine-grained response policies.
- Governance teams visualize the permissions and tools inherited by service tokens to prevent overprivileged agent access.
Specifications
- Deployment
- Cloud-based API integration
- Platforms
- Vertex AI, Amazon Bedrock, Microsoft Copilot
- Integrations
- Workday, Salesforce, ServiceNow
- Supported Platforms
- Vertex AI, Amazon Bedrock, Microsoft Copilot
- SaaS Integrations
- Workday, Salesforce, ServiceNow
- Core Engine
- Attune AI
- Product Stage
- Design Partner Stage
- Primary Controls
- Suspend run, Revoke credential, End session
How it works
- Inventory and Ownership: The platform discovers and inventories all connected agents across AI platforms and identifies the people responsible for them.
- Access Mapping: Abnormal maps the credentials, tools, and sensitive resources that each agent is permitted to access.
- Behavioral Baselining: The system establishes a behavioral baseline for each agent to distinguish between legitimate tasks and rogue actions.
- Containment and Review: When deviations are detected, the platform applies policies to suspend runs or revoke credentials to contain the threat.
FAQ
What is the current status of Abnormal AI Agent Security?
The product is currently in the Design Partner stage, where selected customers evaluate it for free and provide feedback to the product team.
How does it prevent agents from going off-task?
Abnormal connects every agent to its owner, access credentials, and a behavioral baseline to identify when an agent acts beyond its assigned task.
Which AI platforms are supported?
It supports agents running on platforms such as Vertex AI, Amazon Bedrock, and Microsoft Copilot.
What actions can be taken when rogue behavior is detected?
Security teams can use supported controls to suspend a run, end a session, or revoke credentials based on fine-grained policies.
Added by AntSpace from public sources (abnormal.ai).