SUMMARY
This article explains how AI agent activity can be surfaced in AWS Security Hub using Vaikora. Vaikora’s capabilities include providing deep visibility into AI agent operations and generating comprehensive audit logs for real-time monitoring, compliance tracking, and threat analysis. By converting agent-level risk signals into ASFF-compliant findings, organizations can monitor AI behavior alongside traditional cloud security signals within a unified SOC workflow.
Traditional security tools often fail with dynamic, non-deterministic AI systems, making specialized tools and real-time visibility essential for effective AI agent security in AWS environments.
AWS Security Hub has a finding for almost everything that matters in your cloud environment. GuardDuty sends network anomalies. Inspector sends vulnerability assessments. Config sends compliance deviations. Macie sends data classification issues. Every one of those services speaks the same language: ASFF, the AWS Security Finding Format.
Your AI agents don’t appear in any of it.
That’s not a gap Security Hub was designed to fill. Security Hub aggregates findings from services that see the infrastructure layer: network traffic, API calls, resource configurations, workload vulnerabilities. AI agents operate above that layer. When an agent gets manipulated via prompt injection and starts calling S3 APIs it shouldn’t touch, CloudTrail logs the S3 call as legitimate. The agent had valid credentials. The action looks authorized. Security Hub has no finding, because no integrated service saw anything wrong. To ensure secure AI operations, every request made by an AI agent—including those involving sensitive information and API keys—must be monitored, scanned, and validated to prevent data leaks and unauthorized access.
Vaikora monitors the agent layer. Every action gets scored for risk, anomaly-detected, and evaluated against policy before the LLM responds. The connector bridges those scores into Security Hub as native ASFF findings so your SOC sees AI agent behavior in the same console, same format, and same workflows as every other security finding. This integration also supports audit and compliance requirements by providing a verifiable record of agent actions.
Introduction to AI Security
AI security has become a critical concern as AI systems are increasingly integrated into business operations across industries. Unlike traditional IT systems, AI systems introduce unique security challenges that require specialized security controls and expertise. Threats such as prompt injection, data exfiltration, and privilege escalation can compromise not only the AI apps themselves but also the sensitive data and infrastructure they access.
To address these risks, organizations must implement robust security rules and policy enforcement tailored to AI workloads. Human oversight remains essential, as AI systems can behave in non-deterministic ways that automated controls may not fully anticipate. Continuous monitoring of AI behavior, combined with proactive threat modeling, helps security teams identify and respond to emerging threats before they impact critical systems. By prioritizing AI security, organizations can maintain trust in their AI systems and ensure the integrity of their data and business processes.
Benefits of AI Agents
AI agents bring significant benefits to organizations seeking to enhance security, efficiency, and scalability. By automating routine monitoring and response tasks, AI agents free up valuable human resources for more strategic initiatives. These agents can provide around-the-clock monitoring of systems and data, enabling rapid detection and response to security threats that might otherwise go unnoticed.
With AI agents in place, organizations can reduce the risk of human error, maintain the integrity of their systems, and ensure compliance with security policies and regulatory requirements. AI agents also help protect sensitive data by continuously monitoring for threats and anomalies, supporting security teams in maintaining a strong security posture. Ultimately, the integration of AI agents allows organizations to respond more effectively to evolving threats while optimizing the use of both human and technological resources.
Why ASFF Format Matters
Security Hub’s value comes from aggregation. When everything speaks ASFF, you can write detection rules that span sources, correlate findings across services, route alerts through consistent playbooks, and measure your security posture in one place. Findings that don’t conform to ASFF either don’t get into Security Hub at all or show up as raw data that doesn’t plug into your existing workflows. Audit logs generated by Vaikora provide real-time monitoring of threats, request analysis, and system activities, enhancing compliance and offering transparency into security operations.
Vaikora-sourced findings are fully ASFF-compliant. Every field is populated correctly:
SchemaVersion
AwsAccountId
CreatedAt
UpdatedAt
Title
Description
Resources
Severity
Types
Vaikora’s capabilities deliver deep visibility into model behavior, active protection against threats, and rigorous audit capabilities, ensuring effective AI security compliance and meeting regulatory requirements.
Severity Mapping
Severity labels use the Security Hub scale: LOW, MEDIUM, HIGH, CRITICAL. Severity normalized scores map directly from Vaikora’s 0-100 risk scale:
| Vaikora Risk Level | Security Hub Severity Label | Normalized Score |
|---|---|---|
|
Critical |
critical |
95 |
|
High |
high |
75 |
|
Medium |
medium |
50 |
|
Low |
low |
30 |
Finding IDs are stable and prefixed with vaikora-{action_id}. Security Hub deduplicates on finding ID, so re-running the Lambda never creates duplicate findings for the same agent action. Finding type is set to Software and Configuration Checks/AI Agent Security, which fits cleanly into the Security Hub finding type taxonomy.
How the Connector Works
Scheduled Ingestion
A Lambda function runs on a schedule via EventBridge Scheduler, every six hours by default. It polls the Vaikora /actions API for new agent actions since the last run, filters to actions that meet the threshold for ingestion (risk level high or critical, anomaly flag set, or confirmed threat detected), maps each action to ASFF format, and pushes to Security Hub via BatchImportFindings. Each request and agent action is logged for audit and compliance purposes, supporting continuous monitoring by capturing all agent activities and unexpected API calls or high-risk actions.
State Management
A DynamoDB table stores the last_run_time timestamp after each successful run. The next invocation picks up from that timestamp. If a run fails because Vaikora is unreachable, the DynamoDB offset doesn’t update, so the next successful run catches everything that was missed. No actions fall through the gap.
Architecture
The architecture is fully serverless. Lambda plus EventBridge plus DynamoDB runs well under $5/month at typical agent volumes. No EC2, no containers, nothing to manage.
Deployment Stack
The entire stack deploys from a single CloudFormation template. It provisions:
Lambda function
IAM execution role
EventBridge Scheduler rule
DynamoDB table
Secrets Manager secret
The IAM role is scoped to least-privilege:
securityhub:BatchImportFindings
secretsmanager:GetSecretValue
dynamodb:GetItem
dynamodb:PutItem
Nothing broader.
What Shows Up in Security Hub
After the first Lambda invocation, open Security Hub and filter findings by:
ProductName = “Vaikora AI Agent Security”
Any high-severity or anomalous agent actions from the polling window appear as findings. These findings provide deep visibility into agent behavior, including user and identity context, which supports effective threat detection and rapid incident response. Continuous monitoring of AI agent behavior during live operations is essential for detecting and blocking threats that may not be visible during static analysis. AI runtime security safeguards AI applications by continuously monitoring AI behavior at inference, enabling real-time threat detection and proactive defense.
Finding Details
Each finding includes:
A human-readable title describing the action and risk level
Description with the agent ID, action type, risk score, and anomaly/threat flags
Severity label and normalized score matching the Vaikora risk assessment
Resource fields identifying the affected agent and AWS account
Finding ID stable enough to track across invocations
For each finding, audit logs are generated to support compliance tracking and provide a verifiable audit trail, ensuring enforcement of security rules in accordance with governance frameworks such as NIST AI RMF.
Workflow Compatibility
Your existing Security Hub detection rules, aggregation configuration, and response playbooks apply to these findings without modification. If you have EventBridge rules that trigger on HIGH or CRITICAL findings, they fire on Vaikora findings too. If you have Security Hub cross-account aggregation set up, findings from member accounts roll up to your administrator account.
Amazon Detective integration works if Detective is enabled and connected to Security Hub. Vaikora findings become part of the investigation graph for any detective session that spans the relevant time window.
Vaikora’s capabilities include providing deep visibility into AI model behavior, active protection against threats, and rigorous compliance tracking, ensuring effective AI runtime security while integrating seamlessly with your existing AWS workflows.
AI Security Best Practices
Securing AI systems requires a comprehensive approach that combines technical controls with human oversight. Organizations should implement robust security controls tailored to the unique risks of AI, such as access management, input validation, and monitoring for behavioral anomalies. Regular threat modeling helps security teams anticipate and mitigate potential attack vectors, while continuous monitoring ensures that emerging threats are detected and addressed in real time.
Human oversight is crucial for reviewing AI decision-making processes and responding to incidents that automated systems may miss. Protecting sensitive data should be a top priority, with strict controls over data access and usage within AI workloads. By fostering a culture of security awareness and providing ongoing training for security teams, organizations can minimize risks and maintain the integrity and confidentiality of their data and systems.
Deployment in Three Steps
Step 1: Subscribe on AWS Marketplace
Search “Vaikora” in AWS Marketplace. Select “Vaikora AI Agent Signals for AWS Security Hub”. Click Subscribe. No upfront commitment.
Step 2: Launch the CloudFormation Stack
From the Marketplace listing, click Launch CloudFormation Stack. You can quickly create and deploy the solution without coding. Enter your Vaikora API key, Agent ID, and AWS region—API keys are required to securely connect Vaikora with AWS and protect sensitive information. The template provisions everything automatically. Takes about two minutes.
Step 3: Verify Findings
Wait for the first EventBridge trigger or invoke the Lambda manually from the console. Open Security Hub, filter by ProductName, and confirm findings appear.
Security Hub must be enabled in your account before deploying. If it’s not active yet:
aws securityhub enable-security-hub
The polling interval is a CloudFormation parameter. Default is six hours. If your agent volume warrants more frequent ingestion, set it to one hour. The Lambda cost difference is negligible.
The Gap This Closes
Security teams running AWS are used to a specific workflow: something anomalous happens, a service generates a finding, it appears in Security Hub, analysts investigate. That workflow has worked well for infrastructure-layer threats, but industry-specific security needs—such as compliance, automation, and secure deployment—require tailored solutions and monitoring practices to address the unique risks of AI agents.
AI agents don’t fit that workflow because no existing AWS service monitors what agents decide. Vaikora does. The connector makes those decisions visible in Security Hub without requiring a new console, new tooling, or changes to how your SOC operates. High-risk agent signals appear in the same place as GuardDuty findings, with the same format, routed through the same detection rules. The point at which AI agents are granted higher autonomy should be carefully managed: start with limited autonomy, progressively increasing it only as controls and monitoring mature. Agents should be granted only the minimum permissions necessary to perform their tasks, following Least-Privilege principles. Security must also be enforced through deterministic external controls at the infrastructure level, outside the agent’s reasoning loop, to ensure robust protection. Adopting a defense-in-depth approach—integrating traditional cloud security with AI-specific controls—helps secure AI agents in AWS environments against evolving threats.
If your organization is deploying AI agents on AWS and running Security Hub as your cloud security posture hub, this is the missing connection.
Future of AI Security
The future of AI security will be defined by the growing adoption of agentic AI systems and autonomous AI agents across diverse environments. As these systems become more capable and widespread, organizations will face new and evolving security challenges that demand advanced security controls and continuous monitoring. Security teams will need to develop expertise in both AI and traditional security domains to effectively respond to complex threats.
Agentic AI systems will require adaptive security measures that can evolve alongside changing threat landscapes and business environments. Human oversight will remain indispensable, ensuring that AI-driven decisions align with organizational policies and ethical standards. Investment in AI security solutions, ongoing development of security teams, and a commitment to balancing innovation with robust security controls will be crucial for organizations aiming to maintain trust and resilience in the age of autonomous AI agents.
Data443 Cybersecurity Integrations
Deploy native integrations to enrich Microsoft Sentinel alerts, reduce investigation time, and automate response in minutes.
Frequently Asked Questions
Why don’t AI agents appear in AWS Security Hub by default?
Security Hub aggregates findings from infrastructure-level services, while AI agents operate at the application layer. Their actions appear legitimate and are not flagged by existing AWS security services.
How does Vaikora integrate with AWS Security Hub?
Vaikora converts agent-level actions into ASFF-compliant findings and sends them to Security Hub using a serverless connector built on Lambda, EventBridge, and DynamoDB.
What is ASFF and why is it important?
ASFF (AWS Security Finding Format) standardizes security findings across AWS services, enabling correlation, automation, and unified SOC workflows.
What types of AI threats become findings?
High-risk actions, anomalies, and confirmed threats detected by Vaikora are converted into Security Hub findings with severity levels and contextual metadata. Vaikora’s threat detection works in real time, monitoring for suspicious behaviors and identifying incidents such as prompt injection or adversarial input. For example, if an attacker attempts to manipulate an AI agent with a crafted prompt designed to bypass security controls, Vaikora detects this threat and generates a corresponding finding.
Does this require changes to existing SOC workflows?
No. Vaikora findings integrate directly into existing Security Hub workflows, dashboards, and response processes.