Your Penetration Tester Never Sleeps: How AWS Security Agent Is Rewriting the Rules of Application Security
A comprehensive guide to AWS Security Agent, the autonomous AI penetration tester that finds, validates, and fixes vulnerabilities across your entire application portfolio.
The Security Bottleneck That Every Engineering Organization Knows Too Well
Here is a scenario that will feel painfully familiar. Your team ships code multiple times a week. Maybe multiple times a day. But your last penetration test? That was six months ago. The security review queue has 40 applications waiting. Your AppSec team of three cannot keep up with 20 development squads. Something will slip through, and everyone knows it.
This is not a failure of talent or effort. It is a structural problem. Development velocity has outpaced security validation for years. A 2025 Checkmarx report found that 81 percent of organizations knowingly deploy vulnerable code to meet delivery deadlines. Over 60 percent of organizations update web applications weekly or more often, while nearly 75 percent test those same applications monthly or less. The math does not work.
AWS Security Agent was built to close that gap. Previewed at re:Invent 2025 and now generally available for on demand penetration testing, it represents a fundamentally different approach to application security: an AI powered frontier agent that proactively secures applications throughout the entire software development lifecycle, from design documents to deployed code, without requiring your team to slow down.
What Exactly Is AWS Security Agent?
AWS Security Agent (now part of AWS Continuum) is what AWS calls a "frontier agent" for application security. Unlike traditional scanning tools that check code against known patterns, this agent reasons about your application. It reads your design documents, understands your architecture, learns your business logic, and then conducts security assessments the way an experienced human penetration tester would.
The agent delivers three core capabilities across the development lifecycle:
1. Design Security Review
Before a single line of code is written, you can upload architectural documents and product specifications. The agent evaluates them against your organizational security requirements and industry standards (AWS Well Architected Framework, NIST CSF, PCI DSS), surfacing compliance gaps and security risks when they are cheapest to fix.
2. Code Security Review
As developers write and submit pull requests, the agent analyzes code for both common vulnerabilities (OWASP Top 10) and your organization specific security policies. It posts findings directly in GitHub, GitLab, or Bitbucket, with remediation guidance inline. It also performs full repository scans that reason about your entire codebase, not just individual files.
3. On Demand Penetration Testing (Generally Available)
This is the headline capability. The agent conducts autonomous penetration testing that operates like a skilled human tester. It maps your attack surface, crafts targeted exploits, validates findings through actual exploitation, and delivers reproducible proof with developer friendly fix recommendations. What traditionally takes weeks now completes in hours.
The Architecture Under the Hood: A Multi Agent System
What makes AWS Security Agent genuinely different from existing DAST or SAST tools is its multi agent architecture. Rather than running a single scanning engine, the system orchestrates a network of specialized AI agents, each optimized for a different aspect of security testing.
How It Works Step by Step
Authentication and Initial Access: The system begins with an intelligent sign in component that handles authentication across diverse application architectures. It uses LLM based reasoning combined with deterministic mechanisms to locate sign in pages, attempt provided credentials, and maintain authenticated sessions.
Baseline Scanning: Following authentication, parallel specialized scanners conduct automated web application security testing. In black box mode, the network scanner generates traffic interactions and identifies candidate vulnerable endpoints. In white box mode, a code scanner additionally performs deep source code analysis.
Multi Phased Exploration: The system uses two approaches in concert:
- Managed Execution runs predefined tasks across major risk categories (XSS, IDOR, privilege escalation, injection attacks) to ensure comprehensive coverage.
- Guided Exploration takes a dynamic, intelligence driven approach. It ingests discovered endpoints, validated findings, and code documentation to reason about application specific attack opportunities, generating a contextual penetration testing plan that evolves based on application responses.
Specialized Agent Swarm: Both exploration approaches dispatch work to specialized worker agents, each configured for specific risk types and equipped with penetration testing toolkits including code executors, web fuzzers, NVD vulnerability database search for CVE intelligence, and vulnerability specific tools.
Validation and Reporting: When agents identify potential vulnerabilities, findings undergo rigorous validation through both deterministic validators and specialized LLM agents that attempt active exploitation. Validated findings receive CVSS severity scores and are synthesized into actionable reports.
Benchmark Performance
On the CVE Bench public benchmark (40 critical severity CVEs from the National Vulnerability Database), AWS Security Agent achieved a 92.5 percent attack success rate with full feedback, 80 percent without guidance (reflecting real world conditions), and 65 percent when using an LLM whose training data predates the benchmark.
Architecture Diagram
The following diagram shows the end to end flow of how AWS Security Agent conducts autonomous penetration testing:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
┌─────────────────────────────────────────────────────────────────────────────┐
│ AWS SECURITY AGENT ARCHITECTURE │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ ┌───────────────────────────────────────────────────────────┐ │
│ │ INPUT SOURCES │ │
│ │ Design Docs | Source Code (GitHub/GitLab/Bitbucket) │ │
│ │ Running Applications (Any Cloud / On Prem / Hybrid) │ │
│ └───────────────────────────┬───────────────────────────────┘ │
│ ▼ │
│ ┌───────────────────────────────────────────────────────────┐ │
│ │ AUTHENTICATION ENGINE │ │
│ │ LLM Reasoning + Deterministic Auth Handler │ │
│ │ (Locates sign in, manages sessions, handles MFA) │ │
│ └───────────────────────────┬───────────────────────────────┘ │
│ ▼ │
│ ┌───────────────────────────────────────────────────────────┐ │
│ │ BASELINE SCANNING LAYER │ │
│ │ ┌─────────────────┐ ┌─────────────────────┐ │ │
│ │ │ Network Scanner │ │ Code Scanner │ │ │
│ │ │ (Black Box) │ │ (White Box) │ │ │
│ │ └─────────────────┘ └─────────────────────┘ │ │
│ └───────────────────────────┬───────────────────────────────┘ │
│ ▼ │
│ ┌───────────────────────────────────────────────────────────┐ │
│ │ MULTI PHASED EXPLORATION │ │
│ │ ┌──────────────────┐ ┌────────────────────────┐ │ │
│ │ │ Managed Execution │ │ Guided Exploration │ │ │
│ │ │ (XSS, IDOR, SQLi, │ │ (Context aware, adapts │ │ │
│ │ │ Priv Escalation) │ │ based on responses) │ │ │
│ │ └────────┬─────────┘ └──────────┬─────────────┘ │ │
│ │ └──────────┬─────────────┘ │ │
│ └──────────────────────┼────────────────────────────────────┘ │
│ ▼ │
│ ┌───────────────────────────────────────────────────────────┐ │
│ │ SPECIALIZED AGENT SWARM │ │
│ │ ┌────────┐ ┌────────┐ ┌────────┐ ┌────────┐ ┌────────┐ │ │
│ │ │ XSS │ │ SQLi │ │ IDOR │ │ Auth │ │ SSRF │ │ │
│ │ │ Agent │ │ Agent │ │ Agent │ │ Agent │ │ Agent │ │ │
│ │ └────────┘ └────────┘ └────────┘ └────────┘ └────────┘ │ │
│ └───────────────────────────┬───────────────────────────────┘ │
│ ▼ │
│ ┌───────────────────────────────────────────────────────────┐ │
│ │ VALIDATION + REPORTING │ │
│ │ Deterministic Validators | LLM Exploitation Agents │ │
│ │ CVSS Scoring | Remediation Guidance | PDF Reports │ │
│ │ ──► GitHub PR Comments | Jira Tickets | IDE Findings │ │
│ └───────────────────────────────────────────────────────────┘ │
│ │
└─────────────────────────────────────────────────────────────────────────────┘Why This Matters: The Case for Autonomous Security Testing
Traditional Penetration Testing Is Broken at Scale
Most organizations limit manual penetration testing to their most critical applications and conduct tests periodically due to time and cost constraints. An external pen test engagement can cost $20,000 to $100,000 per application. With a portfolio of 50 or 100 applications, comprehensive testing is simply not financially viable.
AWS Security Agent flips this equation. At $50 per task hour with a two month free trial, organizations can expand penetration testing across their entire portfolio. The always on nature means you can test after every significant deployment rather than once a quarter.
It Finds What Pattern Matching Tools Miss
Static analysis tools (SAST) look for known vulnerability patterns. Dynamic analysis tools (DAST) probe running applications without understanding context. Both are valuable but limited.
Consider a real example from the AWS Security Blog: A stored procedure had a SQL injection vulnerability. A traditional SAST tool would flag the specific EXECUTE IMMEDIATE call. AWS Security Agent went deeper. It identified that the central validation function does not block single quotes in any of its five regex profiles, listed all five profiles by name, explained why single quotes matter for the specific database engine, and noted that another stored procedure skips the validation function entirely. Instead of a point fix on one call site, the finding led to a comprehensive remediation of a systemic gap.
Validated Findings Mean Less Noise
One of the most frustrating aspects of automated security tools is false positives. AWS Security Agent addresses this head on. Every finding goes through independent validation where a separate agent re reads the source code and traces the full attack chain. The simulated validation capability deploys your application in a sandbox environment to dynamically confirm whether discovered vulnerabilities are actually exploitable.
How AWS Security Agent Helps Day to Day Productivity
For Developers
Security findings show up directly in pull requests with specific code fix suggestions. No context switching. No waiting for a security review meeting. The Kiro power and Claude Code plugin integration means developers can trigger code reviews, generate threat models, and remediate findings directly from their IDE or CLI.
For Application Security Teams
Design reviews that previously consumed hours of manual effort become focused analyses. Security requirements are defined once and automatically enforced across every application. Penetration testing transforms from a periodic bottleneck into an on demand capability.
For Engineering Leadership
The CI/CD pipeline integration means security testing becomes part of the deployment workflow, not a separate process. Trigger automated tests from your pipelines, integrate results into existing tooling, and validate security at every deployment.
Real World Use Cases
Securing APIs and Web Applications at Scale
An organization maintaining 200 microservices can run penetration tests across all of them. The agent tests against OWASP Top 10 vulnerabilities but goes beyond: it crafts application specific attack chains that combine multiple vulnerability types.
Pre Launch Security Validation
Before any major release, teams can trigger a comprehensive penetration test and receive validated findings within hours rather than scheduling an engagement weeks in advance.
Inherited Codebase Assessment
When acquiring a company or integrating open source code, the full repository scan builds a security model from scratch without needing institutional knowledge of the codebase.
Continuous Compliance Validation
For organizations in regulated industries, the managed compliance packs (AWS WAF, NIST CSF, PCI DSS) and custom organizational requirements mean every design document and code change is automatically evaluated against your compliance posture.
Multi and Hybrid Cloud Security
AWS Security Agent can test any application, whether it runs on AWS, on premises, in a hybrid environment, or in other cloud providers. Once enabled, point it at any accessible endpoint and it conducts comprehensive testing regardless of where the application is hosted.
On Premises and Multi Cloud Connectivity
This is where AWS Security Agent stands apart from many AWS services that only work within the AWS ecosystem. The agent's penetration testing capability is cloud agnostic by design.
On Premises Applications: If your application runs in a private data center, the agent can test it as long as it has network access to the endpoint. Through cross account VPC configurations and private connectivity options, you can expose internal applications to the agent without making them publicly accessible.
Multi Cloud Deployments: Organizations running workloads across AWS, Azure, and GCP can point the Security Agent at any HTTP/HTTPS endpoint regardless of where it is hosted. The agent does not care whether the application runs on EC2, Azure App Service, or Google Cloud Run.
Hybrid Architectures: For enterprises with applications that span multiple environments, the agent tests the complete attack surface as a unified target. It does not need separate configurations for each cloud provider.
Source Code Regardless of Platform: The code security review capability integrates with GitHub, GitLab, and Bitbucket, all of which support self hosted (on premises) deployments.
Practical Impact: For organizations managing a portfolio of applications across legacy data centers, AWS, and other clouds, this means one security testing platform covers everything. One agent, one set of policies, one reporting format, complete coverage.
Enterprise Context: Making Security Scale Across Large Engineering Organizations
For large enterprises operating at scale with distributed engineering teams, the security challenge is not just technical. It is organizational. A small centralized AppSec team cannot personally review the output of hundreds of developers moving at modern velocity.
Consistent Security Enforcement Across Teams: Security requirements are defined once centrally and applied uniformly across all geographies and teams.
Scaling Penetration Testing Across the Portfolio: At $50 per task hour, the calculus changes entirely. You can afford to test everything, not just the crown jewels.
Reducing the Mean Time to Remediation: The full loop from discovery to validated finding to developer actionable fix recommendation happens in the same workflow.
Enabling Developer Self Service: Developers can trigger scans, get findings with remediation, and fix issues themselves. The AppSec team shifts from reviewing everything to configuring policy and handling truly complex issues.
Integration Points and Ecosystem
- Source Control: GitHub (including Enterprise), GitLab, and Bitbucket (both SaaS and self hosted)
- Documentation: Confluence integration for referencing existing documentation as security context
- CI/CD Pipelines: Full API and SDK support for triggering tests from any pipeline
- IDE Integration: Kiro power, Claude Code plugin, and open MCP integration for any AI powered IDE
- AWS Infrastructure: Cross account VPC support, IAM integration, CloudWatch for test logs, KMS for encryption
- AWS Continuum: Continuous vulnerability prioritization, sandbox validation, and automated remediation workflows
- Compliance Frameworks: Managed packs for AWS Well Architected Framework, NIST CSF, PCI DSS, and custom requirements
Pricing and Free Trial
$50.00 per task hour, metered per second. No upfront costs, no minimum fees, no long term commitments.
Because multiple tasks run concurrently during a pen test (the specialized agent swarm working in parallel), total billable task hours typically exceed actual wall clock duration. A pen test finishing in 1 hour of real time might consume 3.5 task hours of cumulative agent work.
Real World Pricing Examples
| Scenario | Wall Clock Time | Cumulative Task Hours | Cost |
|---|---|---|---|
| Simple API test | ~1 hour | 3.46 hours | $173.13 |
| Monthly e-commerce app test | ~4 hours | 24 hours | $1,200.00 |
| Complex enterprise SaaS pre-release | ~9.5 hours | 31.26 hours | $1,562.92 |
Free Trial
New customers receive a 2 month free trial with up to 400 pentesting task hours per month. Design review, code review, and threat modeling are currently free and not billed (up to 200 design reviews and 1,000 code reviews per account per month).
Cost Comparison
| Testing Approach | Typical Cost Per Application | Frequency |
|---|---|---|
| Manual penetration test (external firm) | $20,000 to $100,000 | Annual or quarterly |
| AWS Security Agent (simple app) | ~$175 to $500 | On demand, as often as needed |
| AWS Security Agent (complex enterprise app) | ~$1,200 to $2,000 | On demand, as often as needed |
Limitations Summary
Key boundaries: the agent uses Kali Linux penetration testing tools that may modify application state, so always run against staging or pre production. It is optimized for web applications and APIs (not mobile or desktop apps). Default limit is 5 concurrent pentest runs per account per region. On demand penetration testing is GA; design review, code review, and threat modeling remain in preview. Source control integration covers GitHub, GitLab, and Bitbucket (Azure DevOps, Perforce, SVN not yet supported).
Available in six AWS Regions: US East (N. Virginia), US West (Oregon), Europe (Ireland), Europe (Frankfurt), Asia Pacific (Sydney), and Asia Pacific (Tokyo).
How It Compares
AWS Security Agent occupies a unique position: neither a traditional scanner nor a manual testing firm, but an autonomous AI agent that combines aspects of both.
vs Snyk: Snyk excels at developer experience for dependency management and SCA. AWS Security Agent provides deeper application level penetration testing with validated exploitability proof. They complement each other well.
vs Veracode: Veracode has decades of enterprise compliance certification. AWS Security Agent offers autonomous reasoning with full lifecycle coverage in one tool at dramatically lower cost per test.
vs Burp Suite Pro: Burp is unmatched for manual red team exercises. AWS Security Agent automates what Burp enables manually. Use both: the agent for broad autonomous coverage, Burp for targeted creative exploitation.
vs OWASP ZAP: ZAP is free and catches low hanging fruit reliably. AWS Security Agent finds business logic flaws and validates exploitability, which ZAP cannot.
The key differentiator: AWS Security Agent is the only single tool covering design review, threat modeling, PR code review, full repository scanning, and runtime penetration testing in one platform.
The Bigger Picture: AWS Continuum
In June 2026, AWS announced that AWS Security Agent is now part of AWS Continuum, a broader AI native security service that continuously discovers, prioritizes, validates, and remediates security risks across the entire software lifecycle.
Continuum adds: a Context Graph that prioritizes findings using environment and business context; Sandbox Validation that confirms exploitability automatically; Automated Remediation with learn and enforce modes; and supervised human in the loop operation that earns greater autonomy as customers grant permission category by category.
Frequently Asked Questions
Q: Does my application need to be hosted on AWS?
No. The agent can test any application accessible via HTTP/HTTPS, regardless of where it is hosted.
No. The agent can test any application accessible via HTTP/HTTPS, regardless of where it is hosted.
Q: Will the agent modify my application or data?
Potentially, yes. It uses Kali Linux penetration testing tools. Always run tests against staging or pre production environments.
Potentially, yes. It uses Kali Linux penetration testing tools. Always run tests against staging or pre production environments.
Q: Does AWS use my code or data for model training?
No. AWS explicitly states the agent does not use customer data for model training or share it with third parties.
No. AWS explicitly states the agent does not use customer data for model training or share it with third parties.
Q: What is the difference between AWS DevOps Agent and AWS Security Agent?
They are complementary. DevOps Agent handles release management, incident investigation, and operational reliability. Security Agent handles threat modeling, penetration testing, and security vulnerability scanning.
They are complementary. DevOps Agent handles release management, incident investigation, and operational reliability. Security Agent handles threat modeling, penetration testing, and security vulnerability scanning.
Q: How long does a penetration test take?
Depends on complexity: simple API tests finish in about 1 hour, medium applications about 4 hours, complex enterprise SaaS about 9.5 hours of wall clock time.
Depends on complexity: simple API tests finish in about 1 hour, medium applications about 4 hours, complex enterprise SaaS about 9.5 hours of wall clock time.
Key Takeaways
- AWS Security Agent is not another scanner. It reasons about your application like a human security researcher, understanding architecture, trust boundaries, data flows, and business logic.
- The economics change everything. At $50 per task hour, organizations can afford to penetration test their entire application portfolio.
- Validation eliminates noise. Every finding is independently validated through actual exploitation attempts.
- It meets developers where they work. Findings appear in pull requests with specific code fixes.
- Coverage spans the full lifecycle. From design documents through code review to deployed application testing, security validation is continuous.
References
- AWS Security Agent Product Page
- AWS Security Agent Features
- AWS Security Agent Pricing
- AWS Security Agent FAQs
- Inside AWS Security Agent: A Multi Agent Architecture for Automated Penetration Testing
- AWS Security Agent On Demand Penetration Testing Now Generally Available
- AWS Security Agent Adds Threat Modeling, Kiro Power and Claude Code Plugin
- AWS Security Agent User Guide
- CVE Bench: Benchmark for AI Penetration Testing
- OWASP Top 10 Web Application Security Risks
Published June 2026
Co-authored by Sekh Mohammad Warish and Nitish Tipparaju
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article