Skip to content

AVATAR — Automated Vulnerability Assessment, Testing, and Autonomous Remediation

  • Product: Automated Vulnerability Assessment, Testing, and Autonomous Remediation
  • Family: DevOps Toolset

Objective

Adopt AWS Security Agent (and eventually AWS Continuum) to provide AI-driven penetration testing, vulnerability validation, and automated remediation across all repositories in the GitHub org — complementing existing Datadog code analysis with exploit-validated findings and context-aware attack chain discovery.

Context

Current state: - Datadog code analysis (SAST, SCA, secrets detection) deployed across the GitHub org — provides broad static coverage and dependency scanning. - AWS GuardDuty deployed across the AWS org — runtime threat detection at the infrastructure layer, no source code integration. - Amazon Inspector deployed across the AWS org — infrastructure and workload vulnerability scanning (EC2, ECR images, Lambda), no application-layer testing. - Gap: No dynamic application security testing (DAST), no penetration testing at scale, no exploit validation of static findings. Datadog flags vulnerabilities but cannot prove exploitability or chain findings into attack paths. Manual pen testing is periodic and limited to critical apps.

AWS Security Agent fills this gap by acting as an autonomous pen tester that ingests source code and documentation, validates exploitability, chains findings, and generates remediation PRs. AWS Continuum (gated preview) extends this into a continuous lifecycle — discover, prioritize, validate, remediate — with graduated autonomy.

Tools In Use

  • AWS Security Agent (GA) — on-demand penetration testing, SAST, DAST, code scanning, remediation PR generation. Now positioned by AWS as part of AWS Continuum rather than a wholly separate product
  • AWS Continuum (gated preview → GA when available) — continuous vulnerability lifecycle management (discover → prioritize → validate → mitigate), with Security Agent as its pen-testing, code-scanning, and threat-modeling component

Under Evaluation

Existing (Complementary)

  • Datadog Code Analysis (SAST, SCA, secrets)
  • AWS GuardDuty (infrastructure threat detection)
  • Amazon Inspector (deployed org-wide — infrastructure and workload vulnerability scanning)
  • GitHub (source control, PR workflow)

Capability Comparison

Capability Datadog Code Security AWS Security Agent AWS Continuum
Static Analysis
Static Application Security Testing (SAST) ✅ (CI + IDE) ✅ (via Security Agent)
Software Composition Analysis (SCA) ✅ (static + runtime)
Infrastructure as Code (IaC) Security
Secrets Detection
Runtime Analysis
Runtime Code Analysis (IAST) ✅ (production, 100% OWASP score)
Dynamic Application Security Testing (DAST) ✅ (via Security Agent)
Autonomous Penetration Testing ✅ (via Security Agent)
Exploit Validation (proof of exploitability) ✅ (via Security Agent)
Attack Chain Discovery ✅ (via Security Agent)
Prioritization & Context
Severity Scoring (CVSS + environment + threat) ✅ (Datadog Severity Score) ✅ (CVSS + app-specific) ✅ (CVSS + business context)
Runtime Reachability Analysis ✅ (via IAST data flow) ✅ (deployed, reachable, production path)
Observability-Correlated Prioritization ✅ (APM/service context)
Continuous Environment-Wide Prioritization ❌ (per-test only)
Remediation
Suggested Code Fixes (PR comments) ✅ (SAST inline suggestions) ✅ (full PR generation) ✅ (via Security Agent)
Automated Remediation PRs ✅ (via Security Agent)
Sandbox-Validated Patches
Guided Remediation Steps
Operational Model
Continuous / Always-On Scanning ✅ (CI + runtime) ❌ (on-demand)
Graduated Autonomy (learn → enforce) N/A N/A
Multi-cloud Pen Testing N/A ✅ (AWS/Azure/GCP/on-prem) ✅ (via Security Agent)
Threat Modeling (STRIDE) ✅ (preview) ✅ (via Security Agent)
Integration
GitHub Integration
CI/CD Pipeline Integration
IDE Integration
Jira Integration
APM / Service Correlation
Availability GA GA Gated Preview

Scope

In scope

  • Connect all GitHub org repositories to AWS Security Agent
  • Configure agent spaces per application/service
  • Enable on-demand penetration testing for deployed applications
  • Enable full repository code scanning
  • Configure automated remediation PR generation
  • Establish workflow for validated findings → developer triage → merge
  • Evaluate and adopt AWS Continuum when generally available
  • Define graduation criteria from learn mode to enforce mode (Continuum)

Out of scope

  • Replacing Datadog code analysis (complementary, not competing)
  • Infrastructure-level vulnerability scanning (covered by GuardDuty and Amazon Inspector)
  • Compliance-driven work (no compliance driver)
  • Non-GitHub source control

Costs

Item Unit Cost Notes
Security Agent pen test $50.00 per task-hour Metered per second, no upfront cost or minimum fee. The ~$1,200-per-app figure is our own estimate (≈24 task-hours) and is not an AWS-published number — validate against real runs during the trial
Security Agent code scanning TBD Preview, pricing not yet published
Continuum TBD Gated preview, pricing not yet published
Free trial $0 2 months for new Security Agent customers, starting at the first pen test run; each trial month includes up to 400 pen-testing task-hours

Risks

Risk Likelihood Impact Mitigation
Continuum remains in gated preview for extended period Medium Low Security Agent GA covers pen testing and code scanning independently
Noisy findings overwhelm developers Medium Medium Start with critical apps, tune context inputs, use Continuum's prioritization when available
Automated remediation PRs introduce regressions Low Medium All PRs go through standard review process; no auto-merge
Overlap with Datadog findings creates duplicate work Medium Low Define triage process: Datadog for broad static coverage, Security Agent for validated/exploitable findings
GitHub org-wide access grant is broad Low Medium Use read-only repository connections; Security Agent doesn't need write access beyond PR creation

Success Criteria

  • All production-deployed repositories connected to Security Agent agent spaces
  • On-demand pen tests runnable against any application endpoint
  • Validated findings (with exploit proof) surfaced for at least top-10 critical applications
  • Automated remediation PRs generated and flowing through standard review
  • Clear delineation documented between Datadog findings and Security Agent findings (who triages what)
  • Continuum access obtained (gated preview or GA) and an adoption decision recorded, or a documented decision to defer — the criterion is met by reaching a decision, not by Continuum reaching GA

Open Questions

  • AWS now describes Security Agent as part of Continuum rather than a standalone product. Does that change the plan's sequencing — adopt Security Agent now and evaluate Continuum later — or does Continuum's gated preview gate access to Security Agent features we want?
  • Does Security Agent's GitHub integration support GitHub Enterprise (our tier), or only github.com?
  • What IAM permissions/roles are required in our AWS org to operate Security Agent?
  • Can Security Agent test applications behind VPN/private endpoints via VPC connector, and do we need that?
  • How do we handle domain ownership validation for internal-only applications?
  • What's the right triage boundary between Datadog findings and Security Agent findings — severity threshold, exploitability, or something else?

References

Decisions

Date Decision Rationale

Changelog

Date Status Change
2026-06-17 draft Initial draft
2026-07-24 planned Status activeplanned: the product is registered but adoption has not started
2026-07-24 planned Review pass: corrected STRIDE threat modeling attribution to AWS Security Agent (preview) per AWS documentation, resolving a contradiction with the capability table; noted that AWS now positions Security Agent as part of Continuum, with a matching open question on sequencing; recorded Amazon Inspector as an existing org-wide tool; cited the $50/task-hour rate and flagged the ~$1,200-per-app figure as our own estimate; made the Continuum evaluation criterion decision-bounded rather than dependent on a GA date; added References