Tech Intelligence.

GitHub added GitHub App support for enterprise-level Copilot seat management APIs, allowing enterprise apps to assign and manage Copilot seats for users and enterprise teams. Access is controlled by a new enterprise fine-grained permission for Apps called Enterprise Copilot license management.

This makes Copilot license provisioning automatable through GitHub Apps instead of only manual or ad hoc workflows. For enterprises, it improves governance over AI license assignment and can reduce provisioning drift.

Review existing enterprise GitHub Apps and update their permissions/workflows if you want to automate Copilot seat lifecycle management.

AWS Billing and Cost Management now lets you build custom billing views from AWS Billing Conductor billing group views and AWS Billing Transfer views, in addition to the existing managed views. You can combine and filter these source views into a single consolidated view for Cost Explorer, Budgets, and the billing console, and share the custom view with other accounts via AWS RAM.

This improves chargeback/showback and downstream billing workflows by reducing the need to manage separate views per customer or billing group. It also makes cost visibility easier to distribute safely across account boundaries without changing underlying access controls.

Review your Billing Conductor and billing transfer setup and consolidate relevant source views into shared custom billing views for finance/showback workflows.

Microsoft Research published threat research showing that an AI agent model can be backdoored by changing fewer than 10 neurons, creating a sleeper agent that behaves normally in evaluations but executes an attacker-controlled command when a trigger string appears. The post recommends defense-in-depth for model supply chain security, including weight diffing against declared base models, behavioral backdoor scanning, using frontier agents as auditors, and recording provenance/audit results in the model registry.

This highlights a real supply-chain and runtime risk for agentic AI: a model can appear safe in testing yet still execute malicious actions once deployed with privileged access. It pushes security teams to treat model weights like code artifacts, with provenance, integrity checks, and continuous monitoring.

Add model weight provenance, diffing, and behavioral scanning to your AI model intake and deployment controls, and restrict agent permissions to the minimum needed.

Azure has introduced Distributed Accelerator for Cloud Storage (DACS) in public preview for AKS. It runs as a distributed read-through cache inside the cluster, using local NVMe disks across nodes to speed up access to cloud storage for AI workloads.

This is an AKS performance feature that changes how pods read data from remote storage, which can affect storage architecture, node sizing, and data locality assumptions. It may also introduce new operational considerations around caching behavior, eviction, and failure domains.

Review whether any AKS workloads could benefit from DACS, and validate the security and data-handling implications before adoption.

Azure Functions now generally supports .NET 11 for apps using the isolated worker model. This lets you target the newer .NET runtime while continuing to use the Azure Functions host and programming model.

This is a runtime/platform update rather than a security control change. It matters mainly if your serverless estate is standardized on newer .NET versions and you need to keep functions aligned with supported runtimes.

Review Azure Functions runtime support in your estate and plan upgrades for any apps that should move to .NET 11.

AWS Security Bulletin 2026-178-AWS reports a local information disclosure issue in Kiro CLI versions 1.16.0 through 2.26.x on macOS and Linux. The CLI stored session transcripts and prompt history with overly permissive default file modes, allowing other local users on the same machine to read content that may include credentials, source code, or other sensitive prompts. The issue is fixed in Kiro CLI 2.27.0.

This is a confidentiality bug affecting developer workstation data, not a cloud control plane issue. It can expose secrets, proprietary code, and incident context if multiple users or shared hosts are involved.

Upgrade Kiro CLI to 2.27.0 immediately and review any affected hosts for exposed session files or leaked secrets.

Amazon EC2 G7 GPU instances are now available in three additional AWS regions: Europe (London), Asia Pacific (Mumbai, Sydney), and South America (Sao Paulo). The update expands the regional footprint for Blackwell-based GPU compute, with support for AI inference, graphics, video transcoding, spatial computing, and GPU-accelerated analytics.

This is primarily a capacity and placement update, not a security or IAM change. It can matter for organizations that need to deploy GPU workloads closer to users or data for latency, residency, or regional resilience reasons.

Review whether any GPU workloads, AMIs, or deployment templates should be expanded to the newly supported regions.

GitHub Copilot has added Gemini 4 Carbon as a generally available model for coding, agentic, and reasoning tasks across IDEs, CLI, the Copilot app, GitHub.com, mobile, and Copilot coding agent. Enterprise and Business admins can now control access to this model via Copilot model policy, and usage is billed under provider list pricing with usage-based billing.

This expands the model choices available for AI-assisted software development and introduces an admin control point for governing model access in enterprise environments. The pricing model also matters for cost tracking and usage governance.

Review Copilot Enterprise model policies and cost controls before enabling Gemini 4 Carbon for teams working on security-sensitive code.

Amazon OpenSearch Service now supports managed cross-Region disaster recovery between two domains in different Regions, with a single API call or console workflow to pair them. It keeps the secondary domain continuously synced, exposes CloudWatch metrics for replication health/lag, and supports switchover/failover with RPO under one minute and RTO under 20 minutes.

This removes the need to build and operate your own replication/failover layer for OpenSearch, which reduces operational complexity for business continuity designs. It also gives you a native DR option with measurable recovery objectives and monitoring.

Review any mission-critical OpenSearch domains for multi-Region DR readiness, especially version 3.7+, KMS, and unsupported instance/storage types, and plan a switchover/failover test.

Azure Front Door Standard and Premium now support gRPC in public preview, allowing HTTP/2-based gRPC traffic to be routed through the edge service. The update indicates method- and service-level routing for gRPC applications, which changes how traffic can be distributed and managed at the Front Door layer.

This expands Azure Front Door beyond traditional HTTP workloads and makes it more viable for modern service-to-service APIs that use gRPC. For cloud platform teams, it can simplify edge routing and global delivery patterns for microservices.

Evaluate whether any gRPC services in Azure should be fronted by Azure Front Door Standard/Premium and test routing, auth, and observability in a non-production environment.

Azure Front Door edge actions are now generally available, allowing lightweight JavaScript to run during request and response processing at the edge. This adds programmable logic for real-time request manipulation or response handling directly in the Front Door data path, reducing dependence on origin-side code for simple edge behaviors.

This changes how teams can implement edge-based traffic handling and response customization without deploying separate compute. For security and platform engineering, any logic executed in the request path needs review for authentication, header handling, input validation, and operational risk.

Review existing Azure Front Door policies and determine whether edge actions introduce any auth, logging, or header-trust risks before adoption.

AWS published CVE-2026-126087 affecting OpenJD Sessions for Python before 0.11.0 and the AWS Deadline Cloud worker agent before 0.33.0 on Linux. A user who can submit jobs to a queue can craft a job that alters PATH and causes the worker agent to execute arbitrary code as the agent OS user, which can expose the AWS credentials attached to that worker.

This is a privilege-escalation and credential-theft issue in a workload execution component, so a compromised job submission path can become a cloud credential compromise. In shared render/compute queue environments, this can turn one authorized user into a path to lateral movement and cloud resource abuse.

Upgrade openjd-sessions to 0.11.0 or later and deadline-cloud-worker-agent to 0.33.0 or later, then review queue permissions and worker IAM role scope.

Azure Route Server now supports dual-stack IPv4/IPv6 route exchange in public preview, allowing hybrid connectivity setups to advertise and learn routes for both address families. Existing IPv4-only deployments are unchanged, so this is an additive networking capability rather than a migration for current customers.

This expands hybrid network design options for environments that are introducing IPv6 without replacing existing IPv4 routing. It can simplify transition planning and reduce the need for separate routing constructs.

Review Azure Route Server usage in dual-stack or IPv6 transition plans and validate route propagation and security controls in a non-production environment.

Ansible community package 15.0.0rc1 is now available as a pre-release on PyPI, bundled with ansible-core 2.22.0 and a curated set of collections. The announcement points to the Ansible 15 unified changelog, the ansible-core 2.22 changelog, and the Ansible 15 porting guide, and notes that release timing follows the core schedule. It also highlights that collection updates can be consumed separately via the generated galaxy requirements file even if you stay on an older ansible-core version.

This matters because pre-releases can expose upcoming module, plugin, and collection behavior changes before the final major release. For platform and security automation, that gives you time to validate playbooks, collection dependencies, and CI/CD pipelines before a production upgrade.

If you maintain Ansible-based automation, test critical playbooks and pinned collections against ansible 15.0.0rc1 in a non-production environment and review the porting guide.

GitHub Copilot CLI’s Microsoft C++ Language Server now ships with its MSBuild compile commands extractor built in and automatically runs it when a project is opened. This removes the previous need for Copilot to discover and invoke a separate configuration skill or for users to supply a compile_commands.json file manually for MSBuild-based C++ projects.

This reduces setup friction and token usage for code intelligence in C++ repos, making language-server features more deterministic for MSBuild builds. It also improves the reliability of developer tooling workflows that depend on accurate compilation databases.

No action needed unless you manage MSBuild C++ repositories using Copilot CLI, in which case validate the new automatic extractor behavior in one project.

Amazon RDS for PostgreSQL released new Extended Support minor versions: 13.23-rds.20260819, 12.22-rds.20260820, and 11.22-rds.20260822. These releases extend access to critical CVE and bug fixes for PostgreSQL major versions that are past standard support, and AWS points to upgrade paths via Blue/Green Deployments, in-place upgrades, or snapshot restore.

Extended Support buys time for teams that cannot immediately complete a major PostgreSQL upgrade, but it also signals dependency on aging database versions. For cloud security and platform teams, this affects vulnerability remediation planning, lifecycle management, and potential support-cost exposure.

Inventory RDS PostgreSQL major versions, confirm which instances are on Extended Support, and prioritize a tested upgrade plan using Blue/Green deployments where possible.

Amazon RDS for MySQL 26.10 is available in the RDS Database Preview Environment, giving you a pre-release build to test. The release includes bug fixes, security patches, READ COMMITTED performance improvements, and new error-log warnings for purge lag and history-list length thresholds.

Preview releases can help teams validate compatibility and performance before a production upgrade, especially when security patches are included. The preview environment has limitations on retention and snapshot portability, so it is suitable for testing rather than production migration planning.

Evaluate MySQL 26.10 in a non-production RDS preview instance and verify application compatibility, monitoring, and backup workflows before any production upgrade plan.

Google Cloud Security is describing how to invoke CodeMender directly from pull requests so it can propose or apply security fixes during review instead of only flagging findings after the fact. The update focuses on integrating automated remediation into the developer workflow, likely around code scanning and PR-based change management.

Shifting from detection-only to fix-in-PR can reduce time-to-remediation for vulnerable code and lower the operational burden on security teams. It also helps standardize remediation in the same workflow developers already use for reviews and merges.

Evaluate whether CodeMender can be integrated into your PR workflow as an automated remediation step for security findings.

AWS API MCP Server had a policy enforcement flaw where operation names were normalized incorrectly before comparing against denyList or elicitList rules. As a result, API operations with consecutive capital letters could bypass intended denial or elicitation controls in versions 0.2.13 through 1.5.0; the fixed version is 1.5.1.

This is a security control bypass in a tool that can mediate AWS API actions, so it can undermine least privilege and approval workflows. If used in automation or agentic workflows, it could let unauthorized AWS operations execute despite policy protections.

Upgrade AWS API MCP Server to 1.5.1 immediately and review any denyList/elicitList policies for operation-name matching assumptions.

AWS Partner Central added a new business planning workflow in the AWS Console for partners to create, collaborate on, and export business plans. The update includes AI-assisted plan readiness scoring, real-time goal tracking, attachments, and APIs/SDKs plus programmatic notifications for workflow integration.

This is a partner-management feature rather than a cloud security or platform control-plane update. The main technical relevance is the addition of APIs/SDKs and notifications, which could be integrated into internal partner operations workflows.

No action needed unless you use AWS Partner Central and want to integrate the new business planning APIs into internal workflows.

GitHub will automatically disable Dependabot alerts on repositories that have had no activity for one year, starting December 8, 2026. Existing alerts are archived rather than deleted, and any push or manually re-enabling alerts restores them and resets the inactivity timer. Repositories covered by an enforced security configuration that enables Dependabot alerts, or enterprises that centrally control the alerts policy, are exempt.

This changes alert continuity for stale repositories, which can create blind spots in vulnerability monitoring if inactive codebases are still deployed or contain long-lived dependencies. The archival behavior reduces noise and backend load, but it also means security teams need governance to ensure dormant repos don’t silently lose protection.

Inventory inactive repositories, identify any that must retain Dependabot coverage, and enforce a security configuration or policy to prevent alerts from being paused where needed.

Amazon S3 Access Grants now supports cross-account discovery and listing of grants with a single ListCallerAccessGrants API call. Using InstanceScope=ALL, you can query your own S3 Access Grants instance plus any instances shared via AWS RAM, and GetDataAccess can automatically locate the owning account and return credentials for the matching grant.

This reduces operational complexity for organizations that federate S3 access across multiple AWS accounts and need a centralized view of data entitlements. It also makes cross-account S3 authorization workflows less error-prone by removing the need to manually query each account.

Evaluate whether your S3 Access Grants discovery and access workflows should switch to InstanceScope=ALL and update any IAM/SCP controls or automation that assume per-account queries.

GitHub updated the default dashboard UI to be more compact and information-dense. Key changes include pinning Feed or Home as the default view, reduced padding in pull request/issue/agent session lists, single-line long titles, a higher default item count for pull requests and issues (six instead of three), and hiding the agent sessions section when none exist.

This is a usability change rather than a platform or security control change. It can reduce time-to-triage by surfacing more work items at once, but it does not alter permissions, auditability, or security behavior.

No action needed unless you want to change your default dashboard view to optimize your triage workflow.

Amazon Bedrock now exposes OpenAI Daybreak Red and Daybreak Blue specialized cybersecurity models through the bedrock-runtime endpoint, with access available via US and Global cross-Region inference profiles. AWS notes that inference data is not used for model training, and access requires separate Daybreak enrollment and AWS approval; regional availability via bedrock-mantle is also referenced in the docs.

This expands the set of security-focused foundation models available inside AWS, which can be used for vulnerability research, detection engineering, and incident response workflows. It also introduces additional identity and access controls considerations because the models are gated behind explicit eligibility and approval.

Review Bedrock IAM controls and approval workflows before enabling Daybreak access, and verify logging/monitoring for any use of the models.