Top line

Microsoft 365 outage: Exchange failure spreads to Teams, Copilot and Safe Links

Microsoft began investigating Exchange Online mailbox failures in the Americas at 11:54 a.m. CT Thursday (EX1550543), and soon said users in any region could be affected. CIO, citing Microsoft's service health dashboard, puts the start at 10:17 a.m. CT. The broader incident, MO1550821, covers Teams calendar and scheduling features, SharePoint features that depend on Exchange, Microsoft 365 Copilot, Defender for Office 365 Safe Links URL checks, and parts of Intune and Purview. Overnight Microsoft extended the email timeout to 48 hours to keep delayed mail from bouncing. At 5:11 a.m. CT Friday it reported improving mailbox access but said "some intermittent issues may continue." Microsoft's explanations of the cause have shifted, and there's no root-cause report yet. The outage also knocked out the Microsoft 365 connectors in Claude and ChatGPT, and both incidents were still open at our cutoff.

Why it matters: Mail, calendars, Copilot and a security control went down together. Check whether Safe Links failed open for your tenant during the outage, and confirm delayed mail arrives within the 48-hour window.

Sources: Microsoft 365 Status (first post) · Microsoft 365 Status (5:11 a.m. CT) · CIO · Claude status

OpenAI activates the Codex "banked reset" Tibo teased

At 3:59 p.m. CT Thursday, Codex lead Thibault Sottiaux posted that "the banked reset has been activated and should land in everyones accounts sometime soon!" That follows his Wednesday hint, which our last two editions reported as not yet delivered. It's a one-time reset that users redeem themselves. It isn't a limit increase or a policy change. He didn't say which plans get it, exactly when it lands, or when it expires, and OpenAI hadn't posted anything official by our cutoff. Under OpenAI's general reset rules, you redeem on desktop, on the web or with /usage in the CLI (not mobile), and only after a covered allowance drops below 80% remaining. OpenAI sets the recipients and expiry for each offer.

Why it matters: Check /usage before buying credits or throttling heavy Codex work. Don't assume every plan or seat gets the reset until it shows up.

Sources: Tibo on X · OpenAI help: usage resets

Gemini 4 Hydrogen is GA, with breaking changes for 3.8 Flash users

Google made gemini-4-hydrogen generally available Thursday at noon CT. It's on the free tier, and the paid tier costs $1.50 per million input tokens and $7.50 per million output tokens. It's also live in Google Search globally, in Antigravity and in Cursor, and it's now the default model for the Antigravity agent in Gemini Managed Agents. Teams moving from 3.8 Flash should test first. Hydrogen rejects the minimal thinking level with an error, defaults to high instead of medium, doesn't support explicit context caching (only automatic implicit caching), ignores temperature, top_p and top_k, and shares one 1,048,576-token window across input, thinking and output. On Google Cloud's Agent Platform it's a "short-term release" with a minimum 45-day retirement notice. In Gemini Enterprise's web and mobile apps it's off by default.

Why it matters: It's a fast, capable default at a mid-tier price, but code that hard-codes minimal, sampling parameters or cached contents will break or behave differently.

Sources: Gemini API changelog · What's new in Gemini 4 Hydrogen · Google Search blog · Google DeepMind

What to do

  • Patch Citrix NetScaler ADC/Gateway and Console. CTX697305 covers 28 CVEs (CVSS v4 up to 9.5), including CVE-2026-120868, which hits default configurations with no workaround. The fixes are 14.1-73.57+ and 13.1-64.39+, with FIPS builds also available. CTX697320 is an unauthenticated auth bypass in NetScaler Console, fixed in 14.1-74.34+. Neither bulletin reports exploitation. CTX697305 · CTX697320
  • Update PostgreSQL and plan off version 14. Versions 18.7, 17.12, 16.16, 15.20 and 14.25 fix 52 security bugs, many of which allow code execution as the database OS user. This is PostgreSQL 14's final release. PostgreSQL release
  • Roll out Edge 157.0.4341.57. Microsoft's update feed lists it as a Critical Stable build carrying most of Chrome 157.0.8103.78's fixes. Microsoft hasn't added it to the Edge security notes yet. Edge update feed · Edge security notes
  • Test before upgrading llama.cpp to v1.0.0. llama-server now refuses tools/MCP on a non-local host without an API key unless you pass --allow-unauthenticated-tools. It also blocks media fetches to private addresses unless you pass --media-allow-private-urls, enforces --cors-origins strictly, and rejects older RPC peers. The maintainer calls these hardening changes, not vulnerability fixes. Release
  • Barracuda's CISA deadline is today, Nov. 13, for CVE-2026-116606 and CVE-2026-111560. CISA added nothing new to the KEV catalog in this window. Barracuda advisory
  • Update: new replacement models for the Dec. 11 GPT-5/o3 shutdown. OpenAI's deprecations page now recommends gpt-6.1-sol for gpt-5, o3, gpt-5-pro and o3-pro, gpt-6.1-terra for gpt-5-mini, and gpt-6-luna for gpt-5-nano. Yesterday's edition gave the earlier gpt-5.6 names. OpenAI deprecations
  • Check for eMudhra (emSign) certificate chains before Nov. 16. MSRC blog

Also notable

  • Thinking Machines: Inkling 2 Small open weights under Apache 2.0. It's a 594B-total, 17B-active model that takes text and images, with a 1M-token context, shipped as NVFP4 and MXFP8 checkpoints. vLLM and SGLang support it, and Together and Cloudflare Workers AI host it at $0.30 input and $1.00 output per million tokens. The model card also links an acceptable-use policy. Announcement · Hugging Face · Cloudflare
  • OpenAI: Daybreak Defense Factory (research preview) and Daybreak Red Team. Defense Factory agents investigate vulnerabilities and draft patches for engineers to review, and Red Team does AI penetration testing of web apps. OpenAI says scans are billed by tokens and "your plan's included usage allowance does not cover these scans." Daybreak blog · Security FAQ · Red Team
  • Claude: two-hour major outage. claude.ai, Claude Code and the Console returned errors from 3:58 to 5:58 a.m. CT Friday. Claude status
  • Claude Meetings beta. Claude can join Google Meet calls for Team and Enterprise plans. It's on by default for Team. Organizations using zero data retention, HIPAA-ready setups or customer-managed keys can't use it during the beta. Claude help
  • Anthropic: self-hosted Claude Managed Agents (private beta). Customers run the agent service on their own infrastructure, and access goes through account teams. Docs
  • GitHub Actions retires ARM32 on Linux and Windows. GitHub changelog

Sources

Microsoft 365 Status · Microsoft 365 Status update · CIO · Claude M365 connector status · Tibo on X · OpenAI usage resets help · Gemini API changelog · What's new in Gemini 4 Hydrogen · Google Search blog · Google DeepMind · CTX697305 · CTX697320 · PostgreSQL release · Edge update feed · Edge security notes · llama.cpp v1.0.0 · Barracuda advisory · OpenAI deprecations · MSRC eMudhra blog · Inkling 2 Small announcement · Inkling 2 Small on Hugging Face · Cloudflare Workers AI · Daybreak blog · Daybreak security FAQ · Daybreak Red Team · Claude status · Claude Meetings help · Self-hosted Managed Agents · GitHub changelog