AWS Continuum brings vulnerability fixes to GA, with approval before autonomous mode

AWS has made Continuum for code vulnerabilities generally available. Its October 22 announcement describes an AI‑assisted workflow that brings together vulnerability findings, application context, tests in an isolated sandbox and proposed patches. The starting mode is Learn, where users approve the remediation fixes the service proposes. [AWS sets out the workflow in its release notice](https://aws.amazon.com/about‑aws/whats‑new/2026/10/aws‑continuum‑for‑code‑vulnerabilities‑is‑now‑generally‑available). For application and security teams, the consequential choice is how much authority to give the system after it has produced a finding. A suggested patch, a test supporting that patch and permission to act on it are separate decisions. Continuum puts those steps into a workflow that can begin with approval and later move to autonomous action within user‑set guardrails.

Continuum builds a context graph from an AWS environment, including resource topology, identity relationships and network reachability. [AWS says](https://aws.amazon.com/about‑aws/whats‑new/2026/10/aws‑continuum‑for‑code‑vulnerabilities‑is‑now‑generally‑available) the service ingests existing vulnerability findings, scans code and uses that context to prioritize findings by business impact. That is a broader input set than an isolated code finding. For a reviewer, the useful question is whether the supplied context represents the affected application and its actual access paths. The announcement describes the information the service uses; it does not establish that a particular customer environment has been mapped completely or that every prioritization decision will be correct.

According to [the release](https://aws.amazon.com/about‑aws/whats‑new/2026/10/aws‑continuum‑for‑code‑vulnerabilities‑is‑now‑generally‑available), Continuum tests findings in an isolated sandbox and provides remediation patches as pull requests, with exploitability proof and a blast‑radius assessment attached. AWS also says the service selects an AI model for each task, without identifying the models or explaining the selection criteria in this notice. Those attachments can give a team more to examine than a proposed code change alone. They still need to be read in the context of the affected application. The notice supplies no independent measurement of detection accuracy, false positives, patch quality or remediation success, so the described proof should not be presented as a demonstrated guarantee for every production deployment.

[AWS describes Learn mode](https://aws.amazon.com/about‑aws/whats‑new/2026/10/aws‑continuum‑for‑code‑vulnerabilities‑is‑now‑generally‑available) as the starting point: Continuum proposes remediation fixes that the user approves. The company describes the user switching to Enforce mode once there is sufficient trust in the system’s precision, allowing it to act autonomously within guardrails the user sets. The transition is therefore a decision about permitted action, rather than simply another way to display findings. Teams evaluating the service should establish what actions they intend to allow and what evidence they need before changing modes. The announcement does not specify the guardrail settings, a numerical trust threshold or an automatic timetable for that transition.

Continuum works alongside Amazon GuardDuty and AWS Security Hub, [AWS says](https://aws.amazon.com/about‑aws/whats‑new/2026/10/aws‑continuum‑for‑code‑vulnerabilities‑is‑now‑generally‑available). The release links to an AWS Region table and the product page, but its body does not enumerate supported Regions or provide prices. Teams need to confirm eligibility and commercial terms for their own deployment rather than infer universal availability from the general‑availability label. A practical evaluation follows the whole path: which findings enter the system, what application context accompanies them, what the sandbox evidence shows, how a patch reaches a pull request and who can authorize the next action. OpenTools’ [coverage of GitHub AI Scan](https://opentools.ai/news/github‑ai‑scan‑pull‑requests‑general‑availability‑limits) examines another AI scanning workflow and the role of pull‑request review. Each product’s findings and permissions need their own review; the existence of a generated fix does not itself settle a deployment decision. *Photo: Ajay Suresh / [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:Amazon_Web_Services_(AWS)_Loft_-_NYC_(48129118457).jpg), [CC BY 2.0](https://creativecommons.org/licenses/by/2.0/). AWS Loft in SoHo, New York City, June 21, 2019; company context, not a depiction of Continuum’s sandbox or its operating location. Native display crop and downscale; no artificial upscaling.*