← All issues

This Week In Email — September 2, 2026

Microsoft is putting new Exchange Online tenants on a 60-day sending leash, spent 12 hours globally offline over a lapsed certificate, and Google is finally pulling the plug on Postmaster Tools v1's reputation buckets. Add an AI that can now send your Gmail without asking first, and a phishing kit that's already popped 4,500 Microsoft 365 accounts past MFA, and you've got a week that's mostly about one company whether you like it or not.

In This Issue

It's a lot. Let's get into it.

Top Stories

Microsoft Puts New Exchange Online Tenants on a 60-Day Sending Probation

Here's the number that should get your attention: 2,200. That's how many external recipients a day a brand-new 100-license Exchange Online tenant will be allowed to send to, starting September 14 — down from the normal ~22,059. Trial tenants get squeezed even harder, from 5,000 recipients per 24 hours to 500.

Microsoft is framing this as anti-abuse. New tenants are the easiest way to spin up a throwaway sending identity, and cutting the Tenant External Recipient Rate Limit (TERRL) for the first 60 days makes that workaround less useful for spammers and BEC crews standing up disposable infrastructure. Worldwide tenants get the new limits September 14; GCC, DoD, and 21Vianet tenants are delayed.

The catch: this doesn't just hit spammers. Agencies, SaaS vendors, and anyone spinning up a fresh M365 tenant for a new client or environment eats the same throttle — for two months, whether they're legitimate or not. If you provision new tenants for clients, budget for the ramp.

Sources: Microsoft Tech Community, EmailExpert

Microsoft 365 Suffers ~12-Hour Global Outage — Traced to a Lapsed Certificate

Downdetector logged roughly 50,000 reports before Microsoft 365 mail send, receive, and auth came back worldwide. The outage started August 31, and Microsoft's own incident notes point to the culprit: an authentication certificate that wasn't renewed on schedule, breaking auth components across Exchange Online. Mailbox connectivity was restored after about 12 hours; search functionality trailed behind.

A forgotten cert renewal, at Microsoft's scale, took down mail for a meaningful chunk of the internet's business correspondence for half a day.

If you run anything transactional or business-critical through M365 as your only path, this is your reminder that "Microsoft's infrastructure" is not a synonym for "resilient." Certificate expiry is the most boring, most preventable failure mode there is — and it just proved it can still take down a hyperscaler.

Sources: TechCrunch, Born City

Google Postmaster Tools v1 Retirement Resumes

Google paused the shutdown of Postmaster Tools v1 around its original October 31, 2025 date. That pause is over. Google is now pulling v1 access domain-by-domain, and this time it's sticking.

The headline change: v2 doesn't carry over the "bad / low / medium / high" IP and domain reputation indicators practitioners have used as a quick-glance health check for years. In their place, senders get v2's compliance status and deliverability-analysis sections — more granular data, but by most accounts less actionable at a glance.

If GPT v1's reputation buckets are baked into your monitoring dashboards or your Monday-morning deliverability triage, go rebuild that now, not when your domain's access disappears without warning. The tool isn't gone — the shortcut is.

Source: Spam Resource

Claude Can Now Send Gmail Messages — Without Approval, If You Let It

Anthropic's Claude can now send, reply to, and forward Gmail messages on a user's behalf. By default it shows a draft and waits for a thumbs-up. On Team and Enterprise plans, the plan owner can turn that confirmation off entirely.

Google's own Gemini, notably, still stops at drafting. It doesn't send autonomously. Anthropic just crossed a line Google hasn't.

The nuance worth knowing: this isn't a jailbreak — it's an intentional product decision, gated behind admin controls. But "AI can send email without a human in the loop" is exactly the sentence that should make deliverability and security teams sit up. Expect a rise in AI-originated outbound mail, with all the tone-mismatch, compliance, and fat-fingered-send risk that implies, and watch what it does to engagement-based reputation signals once it's happening at volume.

Source: 9to5Google

Deliverability & Authentication

Twilio SendGrid: 10–15% Spike in Microsoft 365 Blocks Across Its Whole IP Estate

Starting around 21

UTC on August 14, SendGrid customers sending to Microsoft-hosted domains saw a broad increase in "550 5.7.511 Access denied, banned sender" rejections — not isolated to a handful of senders, but across SendGrid's entire IP range. It resolved August 17.

Shared-IP reputation at Microsoft doesn't move slowly or transparently, and this is the proof: a platform-wide block that had nothing to do with any individual sender's practices and everything to do with M365's opaque reputation math on a shared pool.

If you're on shared IPs and send meaningfully to M365 domains, this is why dedicated IPs and M365-specific monitoring keep coming up in every deliverability audit. On shared infrastructure, you're not just responsible for your own reputation — you're exposed to everyone else's.

Sources: EmailExpert, SendGrid Ops on X

Gmail's Verified Political Sender Program Returns September 8

Gmail is relaunching its Verified Political Sender Program, letting FEC-registered (or state/local equivalent) political committees apply to bypass Gmail's spam filter — provided they authenticate via SPF and DKIM, follow Gmail's standard bulk-sender rules, and keep a 14-day complaint rate under 0.3%, verified through the nonprofit Campaign Verify.

This is a rerun, not a new idea. Google first floated a version of this back in 2022, and it was controversial then for the same reason it'll be controversial now: a carve-out that lets one category of bulk sender skip the filter everyone else has to earn their way past.

If you're not in political email, this doesn't touch you directly. If you are, or you advise campaigns on deliverability, September 8 is the registration deadline that actually matters this cycle.

Sources: Spam Resource, Engadget

Security & Anti-Abuse

Mirage2FA Phishing-as-a-Service Hits 3,500+ Orgs, Hijacking M365 Sessions Past MFA

Mirage2FA is a commercial adversary-in-the-middle (AiTM) kit that's compromised an estimated 4,500 accounts across roughly 3,500 organizations, mostly in the US. It routes victims through SVG, HTML, or XHTML attachments into a fake Microsoft login proxy that captures username, password, and — the part that matters — the live 2FA session cookie.

That last piece is what makes AiTM different from garden-variety credential phishing. It doesn't just steal your password. It steals the authenticated session, MFA and all, in real time.

"We have MFA" stopped being a complete answer to account-takeover-driven BEC a while ago. This is the latest, well-quantified proof. If your security stack still treats MFA as the finish line rather than one layer, this is the week to revisit that assumption.

Source: The Hacker News

Events & Community

Links Worth Your Time

That's the week. If something is wrong, reply and tell me — I read every response.

— John

This Week In Email — thisweekin.email

Enjoyed this issue?

Subscribe to This Week in Email and get future issues delivered to your inbox.