Top line

nginx fixes HTTP/2 flaws that can lead to remote code execution

F5 released nginx 1.31.9 (mainline) and 1.30.7 (stable) Wednesday morning, fixing a batch of vulnerabilities. The most serious, CVE-2026-123007, is a memory-corruption bug in HTTP/2 flow control. Its only precondition is HTTP/2 enabled on a listening port, and it affects 1.9.5 through 1.31.8. Also fixed: CVE-2026-123633, an HTTP/2 use-after-free when proxying, plus HTTP/3, resolver and mail-proxy bugs. CERT-EU says all five can lead to unauthenticated code execution as the nginx worker, and that researchers have demonstrated CVE-2026-123007 but no in-the-wild exploitation has been reported. NGINX Plus fixes are 37.0.8.1 and 37.3.0.1. Rapid7 lists no fix for R34 through R36.

Rapid7, which also reported CVE-2026-123007, says its code-execution exploit needs about 128 GB of HTTP/2 traffic and a specific x86-64 glibc copy routine. Bandwidth and timeouts limit exploitation over the internet. Ubuntu shipped patched packages overnight. Release 1.31.9 also starts rejecting header lines without a colon, which can break clients that send malformed headers.

Why it matters: nginx sits in front of most web stacks, ingress controllers and appliances. Patch internet-facing instances first. If you cannot, CERT-EU says turning off HTTP/2 on exposed listeners mitigates both HTTP/2 bugs. Check vendor appliances and container images that bundle nginx.

Sources: nginx security advisories · nginx 1.31.9 changes · CERT-EU advisory · Rapid7 analysis · Ubuntu USN-9159-1

Google releases Gemini 4 Neon between Carbon and Hydrogen, at $3/$15

Google launched Gemini 4 Neon Wednesday afternoon for complex coding and analysis. It is generally available on the paid Gemini API tier only, at $3 per million input tokens and $15 per million output tokens (thinking tokens included), and $0.15 for cached input. Batch and Flex are $1.50/$7.50. It has a 1M-token context window and four thinking levels; the default is high. For comparison, Carbon is $2/$10 through November 30 and $4/$20 from December 1.

Migrating from another Gemini 4 model needs no request changes, though Neon defaults to a higher thinking level than Carbon. Coming from Gemini 3, explicit context caching is not supported, so remove cached_content fields. Neon is in Antigravity for paid Google AI Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra. GitHub Copilot has it for Pro+, Max, Business and Enterprise at list price, and it is on by default unless admins have turned off automatic model enablement. Cursor added it too.

Why it matters: For Carbon workloads, Neon is worth testing before Carbon's price doubles on December 1. Copilot admins who don't want new models turned on automatically should check their model policy.

Sources: Google announcement · What's new in Gemini 4 Neon · Gemini API pricing · Antigravity blog · GitHub changelog · Cursor's post

What to do

  • Patch LiteLLM and the other new KEV entries. CISA added LiteLLM CVE-2026-40217, a sandbox escape in the /guardrails/test_custom_code endpoint that lets someone with a proxy-admin credential run code in the proxy, which runs as root in the default Docker image. The advisory says upgrade to 1.83.11 or block that endpoint; the federal due date is December 2. Synology DSM, Apache Axis and REDCap were added with a November 21 deadline. CISA alert · LiteLLM advisory
  • Effective today: the BIS remote-access chip rule. It was published in today's Federal Register (91 FR 74970) and is in effect. Data centers already supplying significant compute to companies headquartered in Macau or Country Group D:5 must get a license within 90 days. Comments are due January 19, 2027. Federal Register
  • Azure OpenAI: o1, o1-pro, o3, o3-pro, o3-mini, o3-deep-research and o4-mini retire today. Microsoft lists gpt-5.6-sol or gpt-5.6-terra as replacements, but they aren't offered for Batch or Regional Standard deployments, and the schedule names alternatives for those. Vertex AI's gemini-3.6-flash also retires today. Azure retirement schedule · Vertex AI model versions
  • Move off Claude Sonnet 4.5 before November 30. It retires on the Claude API that day, including the undated claude-sonnet-4-5 alias; the replacement is claude-sonnet-5-5. Update Claude Code and any pinned subagent models. Anthropic deprecations

Also notable

  • GitHub Copilot: eight more services can connect through GitHub-verified MCP connectors, including Supabase, Box and Calendly, in public preview on all plans. Business and Enterprise admins must enable the policy. GitHub changelog
  • Claude Code 2.1.328: auto mode for Opus 4.6 and Sonnet 4.6 on Bedrock, Google Cloud, Foundry and the Claude apps gateway. Interactive sessions with no permission mode set now start in it. Release notes
  • OpenAI API: discounted background completion windows (catch-up from Nov. 17). In the Responses API beta, a two-hour target cuts token rates 30% and an eight-hour target 40%, on GPT-6.2 Astra and Sol, GPT-6.1 Terra and GPT-6 Luna. Separately, EU data-residency use of GPT-5.2, 5.4 and 5.4 mini on the fast or priority tier ends January 11, 2027. API changelog · Deprecations
  • DeepSeek: server-side search and web fetch are no longer billed per call (catch-up). Since Nov. 18 at 12:00 Beijing time, they cost only the tokens their results add. DeepSeek pricing
  • Microsoft Edge still hasn't shipped Chrome 158's security fixes. Its notes still say a fix is in progress, and 157.0.4341.57 is the latest stable listed. Edge security notes

Sources

nginx security advisories · nginx 1.31.9 changes · CERT-EU advisory · Rapid7 analysis · Ubuntu USN-9159-1 · Google announcement · What's new in Gemini 4 Neon · Gemini API pricing · Antigravity blog · GitHub changelog: Gemini 4 Neon · Cursor's post · CISA alert · LiteLLM advisory · Federal Register · Azure retirement schedule · Vertex AI model versions · Anthropic deprecations · GitHub changelog: connectors · Claude Code 2.1.328 · OpenAI API changelog · OpenAI deprecations · DeepSeek pricing · Edge security notes