Artificial intelligence news.
Straight from the original announcements.

Wiz Green Agent maps fixes for a critical cloud attack path

November 13, 2026, 18:05By the AI News Bites Editorial TeamSource: Wiz
Wiz Green Agent maps fixes for a critical cloud attack path
Image: Wiz

Wiz has detailed an investigation by Green Agent into an internet-facing Kubernetes container with a live vulnerability and access to sensitive customer data. The investigation mapped the attack path, identified immediate cloud mitigations and linked permanent fixes to the teams responsible for the code.

Key points
  • Green Agent investigated a critical Issue involving an internet-facing AWS EKS workload.
  • CVE-2026-111497 was confirmed as actively loaded in the running Node.js process.
  • The workload could reach S3 buckets containing more than 5,000 unmasked records.
  • Disabling token automount and reducing IAM permissions offered immediate cloud mitigations.
  • Green Agent linked permanent fixes to Dockerfile, Terraform and GitOps owners.

Five risks converged on one production workload

According to Wiz, the Issue centered on a single Kubernetes deployment in an AWS EKS cluster. Its Security Graph connected five risk types: direct internet exposure, the known initial-access vulnerability CVE-2026-111497, a Kubernetes pod mounting a service account token, an IAM role with access to cloud storage, and sensitive customer records in the accessible buckets.

That combination created several possible points for breaking the attack chain. Security teams could remove public exposure, patch the container image, unmount the service account token, reduce the IAM role's permissions or purge the sensitive data. Green Agent first validated the conditions behind each risk before recommending a response.

Green Agent confirmed a live route and a viable exploit

The investigation traced traffic from 0.0.0.0/0 through an AWS Internet Gateway, an AWS Network Load Balancer and the Traefik Ingress controller into the application's main container. Both HTTP on port 80 and HTTPS on port 443 returned unauthenticated 2XX responses, while Wiz found no Web Application Firewall or firewall rules along the route.

Because the service was an intentionally public production frontend, deleting the ingress or blocking the load balancer would have caused a total production outage. Wiz says this ruled out a simple network shutdown and pointed instead toward measures that could preserve uptime, such as authentication middleware.

Green Agent also investigated CVE-2026-111497, identified in the post as “Broccoli2Shell.” Wiz describes it as a critical remote code execution flaw in the Brotli decoder bundled with Node.js and listed in CISA's Known Exploited Vulnerabilities catalog. Wiz Sensor confirmed that vulnerable Brotli libraries were loaded in memory by the running Node.js process. Vendor patches, including Node.js 22.23.4 and libbrotli 1.2.1, were available, although releasing a new image through testing and CI/CD would take time.

Cloud mitigations could act before a code deployment

Green Agent found several ways to cut the cloud portion of the attack path immediately. It confirmed that the Kubernetes service account projected cloud credentials into the container through IRSA, making disabling token automount on the pod spec an immediate mitigation. It also found s3:GetObject and wildcard s3:* permissions across three S3 buckets, creating an opportunity to reduce the IAM policy to least privilege.

Queries of those buckets found more than 5,000 unmasked records containing credit cards and PII. Wiz identified purging or masking data at rest as another remediation option. These measures could reduce the risk while developers worked on the permanent code fix.

For longer-term remediation, Green Agent's Code Subagent traced the container image to its Dockerfile, the IAM role to Terraform definitions and the workload to GitOps deployment manifests. Commit history and metadata connected the proposed changes with the application developer, service owner and infrastructure lead.

The verdict combined urgency with business continuity

Wiz says Green Agent classified the Issue as Critical after considering the workload's High Business Impact production projects, the active path to more than 5,000 sensitive records and historical evidence that 100% of similar sampled issues in the subscription had been remediated, with a 0% rejection rate.

It recommended both Mitigate and Fix plans. The former covered immediate actions such as disabling service account token automount, while the latter included updating the Dockerfile's base image and narrowing Terraform IAM policies. Green Agent rejected deleting the ingress because it would take down the public production frontend, and assigned each plan to a verified owner for resolution through a one-click pull request or instructions to a coding agent.

The example shows the role Wiz is giving Green Agent: combining exposure analysis, runtime evidence, cloud configuration, data access and code ownership in one investigation. For security teams, the practical benefit described by Wiz is the ability to act on immediate mitigations without waiting for a full image release, while preserving a path to permanent fixes that will not be overwritten by a later CI/CD build or Terraform apply.

Related news

For developers
For developersNovember 13, 2026, 17:14

Open Agentic Web Challenge offers $13,000 for interoperable AI agents

A global group of technology, research, standards and civil society organizations has launched the Open Agentic Web Challenge, offering $13,000 for open and…

Wiz outlines AI-era approach to fixing application security risks
Products and toolsNovember 12, 2026, 19:32

Wiz outlines AI-era approach to fixing application security risks

Wiz says its Code platform can help application security teams move from vulnerability discovery to remediation by mapping code ownership, generating tailored fixes and…

Products and tools
Products and toolsNovember 12, 2026, 19:21

RemotePC Adds MCP Integration for AI-Assisted Computer Management

RemotePC has introduced an integration that lets customers use ChatGPT, Claude and other compatible AI assistants to monitor remote computers and perform supported…