Wiz Research and other teams disclosed two critical RCE vulnerabilities in NGINX, alongside F5's security release on Nov 18. Google introduced Gemini 4 Neon, now available on GitHub Copilot.
International

AI & models
Gemini 4 Neon launches, available on Copilot
Google introduced Gemini 4 Neon, a model with a 1 million token context window. Pricing is $3 per million input tokens and $15 per million output tokens. The model is available on GitHub Copilot for Copilot Pro+, Max, Business and Enterprise users, rolling out gradually.
What developers should do: Select Gemini 4 Neon in the GitHub Copilot model picker; Business and Enterprise administrators manage access through model policies.
Sources: blog.google GitHub Changelog

Developer tools
VS Code 1.147 adds Send Later and Drafts
Visual Studio Code 1.147 is out with several experimental features for the Agents window. Send Later lets you schedule a local agent request to be sent later, while Drafts keeps multiple unsent requests along with their context and configuration. Send Later only works with sessions on This Computer and requires automations to be enabled, and the local agent host must be running at the scheduled time.
What developers should do: Try Send Later by enabling the sessions.sendLater.enabled setting in the Agents window.
Sources: VS Code

Cloud, DevOps & infrastructure
AWS Lambda adds GPU support for Event Functions
AWS Lambda now supports EC2 G-series GPU instances for Event Functions running on Lambda Managed Instances. GPU functions are deployed as OCI container images and support extended timeouts of up to 90 minutes for asynchronous invocations. The feature is optimized for inference rather than training, so it suits self-built models better than foundation models already available on Amazon Bedrock.
What developers should do: Request a G-series instance quota increase before you start, as new accounts may have a quota of 0.
Sources: AWS News Blog

Cloud, DevOps & infrastructure
Kubernetes plans to remove the IPVS backend
Kubernetes SIG Network plans to remove the kube-proxy ipvs backend in Kubernetes 1.40 (Nov 17). The backend has been deprecated since Kubernetes 1.35 and still works with a warning. ipvs users need to plan a move to the nftables or iptables backend.
What developers should do: Switch your kube-proxy configuration from ipvs to nftables or iptables; if switching to iptables, reboot each node instead of just restarting kube-proxy.
Sources: Kubernetes Blog

Security
Two critical RCE flaws in NGINX
Wiz Research and other teams disclosed two remote code execution vulnerabilities in NGINX, CVE-2026-123633 and CVE-2026-123007, announced by F5 alongside its Nov 18 security release. Both have a CVSS 4.0 score of 9.2 and both require HTTP/2 to be enabled on the affected NGINX server. CVE-2026-123007 affects versions going back to NGINX 1.9.5 from 2015, so anyone running NGINX with HTTP/2 should check immediately.
What developers should do: Update NGINX to the latest version with the security patches, or temporarily disable HTTP/2 if you cannot upgrade yet.
Sources: wiz.io

Web, mobile & platforms
Jetpack Compose 1.13 and Material 3 1.5.0 go stable
The November 2026 Jetpack Compose release is stable at version 1.13, adding polygon shapes, progressive blur and configurable blur for Dialogs and Popups. Material 3 1.5.0 is also stable, while the Compose Material 2 library is deprecated and no longer actively maintained. Upgrading to Compose BOM 2026.11.01 requires Kotlin 2.2 or higher and compileSdk 37.1.
What developers should do: Upgrade Compose BOM to 2026.11.01 and migrate from Material 2 to Material 3.
Sources: Android Developers
Japan

AI & models
Mercari improves search suggestions with embeddings
Mercari Engineering shared how it improved search suggestions on product pages using Item-to-Query Search with Domain-Aware Ruri, a Japanese embedding model fine-tuned on Mercari's search logs, and Reranking with a CTR prediction model (Nov 16). The fine-tuned 32-dimension embedding outperformed the original 768-dimension one in offline evaluation. In A/B tests, purchases via search suggestions rose 17.57% thanks to Item-to-Query Search and 5.09% thanks to the Reranker.
Sources: Mercari Engineering
Vietnam

Cloud, DevOps & infrastructure
Green data center standard proposed
At the 2026 Workers for the Environment Forum, Assoc. Prof. Dr. Nguyen Dinh Tho, Deputy Director of the Institute of Strategy and Policy on Agriculture and Environment, proposed that Vietnam develop a green data center standard with PUE and WUE metrics. For newly built data centers, he aims for a PUE of no more than 1.3 and a WUE below 0.5 liters/kWh; these are recommended research thresholds, not a currently binding standard. He said power density per rack with AI server clusters could rise to 30-100 kW.
Sources: VnExpress Số hóa
Impact assessment
Toratech Blog's assessment of each story above, not news.
| Story | Affects | Impact | Level |
|---|---|---|---|
| Gemini 4 Neon launches, available on Copilot | GitHub Copilot Pro+, Max, Business and Enterprise users | A model with a 1 million token context window is now available in Copilot's model picker. Pricing of $3 per million input tokens and $15 per million output tokens may be a factor when comparing costs. | Medium impactShort term |
| VS Code 1.147 adds Send Later and Drafts | Users of the Agents window in VS Code 1.147 | Local agent requests can be scheduled and multiple drafts can be kept. Send Later only works with This Computer sessions, requires automations to be enabled, and the local agent host must be running at the scheduled time. | Low impactShort term |
| AWS Lambda adds GPU support for Event Functions | Teams running self-built model inference on AWS Lambda | Event Functions run on G-series GPUs with a timeout of up to 90 minutes for asynchronous invocations. New accounts may have a quota of 0, so you may need to request a quota increase before trying it. | Medium impactShort term |
| Kubernetes plans to remove the IPVS backend | Teams operating Kubernetes clusters that use the kube-proxy ipvs backend | The ipvs backend is expected to be removed in Kubernetes 1.40 and has been marked deprecated since 1.35. Clusters still using ipvs may need to migrate to nftables or iptables. | Medium impactLong term |
| Two critical RCE flaws in NGINX | Teams operating NGINX servers with HTTP/2 enabled | Two RCE vulnerabilities rated CVSS 4.0 9.2 may be exploited on servers with HTTP/2 enabled. CVE-2026-123007 affects versions from NGINX 1.9.5, so the range of affected versions is very broad. | High impactHappening now |
| Jetpack Compose 1.13 and Material 3 1.5.0 go stable | Android app development teams using Jetpack Compose and Material 2 | Compose BOM 2026.11.01 requires Kotlin 2.2 or later and compileSdk 37.1, so upgrading may force a toolchain change. Material 2 is marked deprecated, so migrating to Material 3 may be necessary. | Medium impactShort term |
| Green data center standard proposed | Organizations building and operating data centers in Vietnam | The thresholds of PUE no higher than 1.3 and WUE below 0.5 L/kWh are only a research recommendation and not yet mandatory. If they become a standard, the design of new data centers may face additional electricity and water constraints. | Low impactLong term |
Toratech's take today
Toratech Blog's analysis of the stories above, not news.
-
NGINX RCE vulnerabilities expose both HTTP/2 configurations and code dating back to 2015
According to the report, the two vulnerabilities CVE-2026-123633 and CVE-2026-123007 both have a CVSS 4.0 score of 9.2, and both require HTTP/2 to be enabled. CVE-2026-123007 affects NGINX from 1.9.5 in 2015, so older servers that have not been upgraded may be within the scope of risk.
Based on: Two critical RCE flaws in NGINX
-
Removing IPVS forces kube-proxy to switch backends, with caveats when moving to iptables
Kubernetes SIG Network plans to remove the ipvs backend in Kubernetes 1.40; this backend has been marked as deprecated since 1.35. Clusters still using ipvs may need to switch to nftables or iptables, and with iptables each node needs to be restarted.
Based on: Kubernetes plans to remove the IPVS backend
-
GPU inference on Lambda opens up more options for self-built models
AWS Lambda supports G-series GPUs for Event Functions on Lambda Managed Instances, with OCI container images and a timeout of up to 90 minutes for asynchronous invocations. According to the report, the feature suits inference for self-built models better than foundation models on Amazon Bedrock. The G-series quota for new accounts may be 0.
Based on: AWS Lambda adds GPU support for Event Functions
This digest was compiled with the help of AI; each item names its sources at the end. Captioned images belong to the credited sources; uncaptioned news cards are made by Toratech Blog.
