AWS Builder Center
AWS Continuum: Bringing Context-Aware Security into Codex, Claude Code, and Kiro

AWS Continuum: Bringing Context-Aware Security into Codex, Claude Code, and Kiro

Explore how AWS Continuum combines AI models, AWS-environment context, and sandbox validation to help developers discover, prioritize, and remediate code vulnerabilities directly within their coding workflows.

AWS Continuum: Bringing Context-Aware Security into Codex, Claude Code, and Kiro

AI coding assistants are rapidly changing how software is designed, written, reviewed, and maintained. They can generate application logic, explain unfamiliar code, create tests, and suggest fixes within seconds.
However, generating code faster introduces an important question:
How can organizations make sure that AI-generated code is secure within the environment where it will actually run?
A code pattern may look safe when reviewed in isolation but become vulnerable when combined with an overly permissive IAM policy, an exposed network path, or an insecure cloud configuration.
AWS is addressing this challenge by partnering with Anthropic and OpenAI to bring AWS Continuum for code vulnerabilities into Anthropic Claude Code, OpenAI Codex, and Kiro.
AWS Continuum for code vulnerabilities is currently available in preview, while the integrations with these coding environments are coming soon.

What is AWS Continuum?

AWS Continuum for code vulnerabilities is an AI-powered security capability designed to discover, prioritize, validate, and help remediate vulnerabilities at machine speed.
Traditional security scanners can identify suspicious code patterns, but they often lack the broader context required to determine whether a finding represents a serious risk in a specific environment.
Continuum adds context from the customer’s AWS environment, including:
  • AWS configurations
  • IAM policies
  • Network topology
  • Exposure surfaces
  • Application and infrastructure relationships
This context allows Continuum to move beyond asking, “Could this code be vulnerable?”
It can help answer the more valuable question:
“Is this vulnerability exploitable in this environment, and how urgently should it be addressed?”

Why context matters in vulnerability management

Security teams frequently receive large volumes of findings from static analysis tools, dependency scanners, cloud-security platforms, and penetration tests.
Not every finding carries the same level of risk.
For example, imagine that two applications contain the same vulnerable code pattern. The first application is isolated inside a private network with tightly controlled permissions. The second is internet-facing and uses a role with broad access to sensitive AWS resources.
Although the code-level finding may appear identical, the business risk is very different.
By evaluating code alongside IAM permissions, network exposure, and AWS configurations, Continuum can help prioritize the vulnerability that presents the greater practical risk.
This has the potential to reduce alert fatigue and allow security teams to focus on vulnerabilities that matter most.

How the developer workflow works

Within Codex, Claude Code, and Kiro, developers will be able to initiate on-demand vulnerability scans.
The expected workflow is:
  1. The coding assistant identifies potential vulnerabilities and sends the findings to AWS Continuum.
  1. Continuum analyzes the findings using the context of the customer’s AWS environment.
  1. It prioritizes the vulnerabilities based on their potential impact and exposure.
  1. Continuum validates findings inside a sandbox.
  1. Contextual security intelligence is returned to the coding assistant.
  1. The assistant adjusts its remediation recommendation based on the validated findings.
This creates a feedback loop between the coding environment and the AWS security context.
Instead of switching between an IDE, scanner, ticketing system, cloud console, and remediation workflow, the developer can receive a security-informed recommendation while working on the code.

From a fragmented process to a continuous outcome

A traditional vulnerability-remediation process may involve several separate stages:
Write → Scan → Triage → Prioritize → Fix → Rescan
These activities may also involve multiple teams, including developers, security engineers, cloud-platform teams, and application owners.
Each handoff introduces time, additional context sharing, and the possibility of delays.
AWS describes the Continuum integration as an effort to collapse these stages into a single outcome: a code suggestion that has already been informed by contextual security analysis and validation.
It does not eliminate the need for secure design, human judgment, code review, or governance. However, it can move meaningful security feedback much closer to the moment when code is created.

The role of the AI harness

One of the most interesting parts of the announcement is AWS’s focus on the AI harness.
A foundation model is only one component of an enterprise AI system. To produce reliable outcomes, the model must be connected to:
  • Tools
  • Workflows
  • Guardrails
  • Memory
  • Organizational context
  • Identity and access controls
  • Observability
  • Validation mechanisms
Together, these components form the AI harness.
AWS Continuum uses an agent-team loop architecture and orchestrates different models for different parts of the vulnerability-management process. It selects models, connects them with the customer’s environment, evaluates findings, and coordinates validation and remediation activities.
This is important because different models may perform better at different security tasks. Rather than expecting one model to handle every stage, Continuum acts as an orchestration layer that can apply the appropriate model and context to each task.

Existing applications versus greenfield development

AWS describes two primary modes of operation.

Existing code

For existing applications, Continuum can help discover, prioritize, validate, and remediate vulnerabilities across the environment.
This can support teams managing large codebases where vulnerability backlogs are difficult to prioritize.

Greenfield code

For new applications, developers will be able to use Continuum through Codex, Claude Code, or Kiro to receive security-validated suggestions within their development environment.
This shifts security further left by introducing contextual feedback before vulnerable code progresses deeper into the delivery pipeline.

Why this matters for DevSecOps

The announcement represents a shift from security tools that merely report problems toward systems that participate directly in software delivery.
Potential advantages include:
  • Earlier identification of security risks
  • Better prioritization using AWS-environment context
  • Fewer disconnected developer and security workflows
  • Sandbox validation before recommendations are returned
  • Faster remediation of high-impact vulnerabilities
  • Reduced manual triage for security teams
  • More consistent security feedback across development environments
For DevSecOps teams, the value is not simply another vulnerability scanner. The value lies in connecting code, cloud context, AI reasoning, validation, and remediation inside a continuous workflow.

Architecture considerations

Organizations evaluating this approach should still consider several important controls:

Human oversight

High-impact remediation should continue to follow established review and approval processes. AI-generated fixes should be reviewed, tested, and monitored like other code changes.

Least-privilege access

Any security tool analyzing cloud resources should receive only the permissions required to perform its intended functions.

Data governance

Teams should understand what code, findings, and environmental context are processed and ensure that their usage aligns with organizational data-handling requirements.

Pipeline integration

Developer-time scanning should complement—not necessarily replace—CI/CD security controls, dependency scanning, infrastructure-as-code validation, and runtime monitoring.

Measurement

Teams should track outcomes such as false-positive rates, mean time to remediation, reopened vulnerabilities, and developer adoption to determine whether the integration is improving security.

Final thoughts

The most significant part of this announcement is not that AI can find vulnerabilities. Frontier models are already becoming increasingly capable of identifying insecure patterns and reasoning through complex attack paths.
The larger development is the combination of:
AI reasoning + AWS context + orchestration + sandbox validation + developer workflow integration
That combination can help transform vulnerability management from a fragmented sequence of scans, tickets, and handoffs into a more continuous and context-aware engineering process.
The future of AI-assisted development will not be defined only by how quickly an assistant can generate code. It will also depend on whether that assistant can produce recommendations that are secure, validated, and appropriate for the environment where the software will operate.
AWS Continuum is an important step in that direction.

Availability

AWS Continuum for code vulnerabilities is currently available in preview through AWS. Access can be requested from the AWS Continuum page.
The integrations with OpenAI Codex, Anthropic Claude Code, and Kiro are coming soon.

Reference

Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article