Gmail Outage Map

Community Discussion

Tips? Frustrations? Share them here. Useful comments include a description of the problem, city and postal code.

Beware of "support numbers" or "recovery" accounts that might be posted below. Make sure to report and downvote those comments. Avoid posting your personal information.

Gmail Issues Reports

Latest outage, problems and issue reports in social media:

• 
  terios_ (@_terios) reported 16 minutes ago

  gmail hit me with "happening soon" for a flight i cancelled like 3 months ago and that **** gave me a heart attack for about 10 seconds because i thought i was somehow still like $1k down

• 
  Ai Agency (@AiAgencyGoat) reported 21 minutes ago

  A 12-person accounting firm was spending 6 hours a day on manual invoice processing. Names. Amounts. Due dates. Line items. All entered by hand. One agent cut that to 25 minutes of review. Here's the complete build. 🧵 The size of this problem: Manual invoice processing costs businesses: → $15-40 per invoice in labor (industry estimate) → 3-5% error rate causing payment delays and disputes → 2-6 hours per day for businesses handling high volume For a business processing 50 invoices/week: → 50 x $25 average = $1,250/week = $5,000/month in labor → Your agent: $1,500/month setup + $800/month retainer → Client saves $3,200/month net — from day one That math closes itself on every discovery call. The complete invoice processing agent: Trigger: → Gmail trigger: new email arrives with PDF attachment at billing inbox → Or: file uploaded to a designated Google Drive folder Stage 1 — PDF extraction: → n8n's PDF node extracts all text from the attachment → Handles multi-page invoices automatically → Passes raw text to the next stage Stage 2 — Structured data extraction (LLM): System prompt instructs the LLM to return JSON only: → vendor_name → invoice_number → invoice_date → due_date → subtotal → tax_amount → total_amount → line_items (array) → payment_terms → extraction_confidence (high/medium/low) Stage 3 — Validation: → IF confidence is low → flag for human review → IF required fields are missing → flag for human review → IF total doesn't match line item sum → flag for human review → All other invoices → proceed automatically Stage 4 — Data output: → Appends new row to Google Sheets with all extracted fields → Or: creates record in accounting software via API (QuickBooks, Xero) → Timestamp and source email logged for audit trail Stage 5 — Notifications: → Slack message to accounts team: invoice details summary → Email reply to sender confirming receipt (optional) → Urgent flag if due date is within 7 days Stage 6 — Exception handling: → Flagged invoices go to a separate "Needs Review" sheet → Slack alert to designated reviewer with the specific issue noted → Reviewer corrects and marks as resolved The numbers clients care about: → Processing time per invoice: 8 seconds vs 8 minutes → Error rate: near zero vs 3-5% manual → Invoices that need human review: 10-15% (unusual formats, poor scan quality) → Hours saved per week: 4-20 depending on volume The expansion upsell: After 60 days, propose: "Now that we have all your invoice data in a structured database, I can add a payment tracking agent that monitors due dates and sends automated reminders to vendors. Want me to scope that?" One agent becomes two. Retainer increases. I cover this full build in my course. Follow and like this post and I'll DM it to you. (Must be followed for DM)

• 
  Batuhan Saygılı (@batuhansygili) reported 23 minutes ago

  @TeamYouTube Someone stole my Gmail/youtube account and added themselves from family options as a parent. When I try to log in with my rescue email and number, I encounter a parental lock. I am 26 years old. Can you help me overcome this problem?

• 
  Anon frog poster (@WebBuildingFrog) reported 24 minutes ago

  @CleansedTweets You can appeal quickly gets given back. I've got over 59 Gmail, had this issue in all of them

• 
  Bill Curry (@Avee8tour) reported 24 minutes ago

  @piersmorgan You think? Didn't answer a ******* question, not even what color is the carpet... So it's pretty clear he has tons to hide. Too bad he's too ******* stupid to not write it down in a diary of in a Gmail email, and then brag how it can't be found through FOIA requests.

• 
  muhammadriyaz (@riyaz0786ahmad) reported 27 minutes ago

  @googleaccount "Sir, I am facing a 'Same Email OTP' loop issue on my Google account [Apna Gmail ID Yahan Likhein]. I don't have access to this email as it is logged out, but Google is sending the verification code to the same email. Please help me recover my account.

• 
  Cyrus (@cyrusasg) reported 28 minutes ago

  some takeaways from the kimi k3 report i think are most relevant for app layer companies that invest in post training: 1. kimi decomposes the agent harness into configurable modules (tool interfaces, system prompts, context management, skills, memories, subagents) and instantiates different configs per task group during RL, because a fixed harness overfits the model to that tool schema and interaction protocol. they randomize because they’re shipping weights into harnesses they don’t control. if you own the harness and it’s identical at train and inference, overfitting to it can be free performance. instead randomize the axis that actually varies at deploy. for many app layer companies that’s the tool schema, since tools are customer-defined and they change. and it means sampling your current customers isn’t enough. the schemas you need to perform don’t necessarily exist yet, so synthetic augmentation is a powerful lever. 2. for assistant tasks they built mock gmail, notion, slack, canvas. core semantics preserved, no external apis or rate limits. on top of that, they added persistent environments over multiple simulated days, dozens of interdependent events across apps, each carrying its own eval criterion, and up to thousands of tool calls per rollout. the takeaway is that if your agent acts on customer systems, the mock is not a testing convenience, it’s the training environment, and its fidelity is a hard ceiling on what RL can teach. the expensive part is the event stream and the state that persists across it. 3. non-verifiable rewards go through a generative RM with tournament-style binary comparisons. but the judge doesnt just pick a winner, it has to read the outputs, generate a rubric for that specific comparison, score both against it, and write those to a scorepad. the pairwise outcome is still the reward, the rubric is scaffolding it has to produce first. two things worth stealing. binary comparison mitigates the calibration problem absolute (ie likert) scoring has, and forcing per-comparison criteria before the verdict keeps it from collapsing into vibes. 4. the anti-reward-hacking design runs on one assumption: the agent optimizes against whatever it can see and touch, so you keep the scoring surface out of reach. in their autonomous execution tasks that means agents never see the verifier. a public verifier returns diagnostic feedback the agent can act on, a hidden verifier scores held-out scenarios and never talks back, and submissions carry a penalty under a limited budget so brute-forcing the feedback loop costs something. at the infra layer, sandboxes get forked so grading runs on an exact clone and can’t perturb the environment the agent is still working in. then they assume it gets through anyway. the hack list is maintained and extended as new strategies show up mid-training. ie for kernels that meant catching cuda graph replay, input caching, precision reduction. web dev zeroes the reward outright if the project fails to build or fakes the artifact instead of implementing it. the takeaway is to try to separate the signal you show the agent from the signal you score it with, and treat reward hacking as an ongoing surface with an owner rather than a bug you fix once.

• 
  🌹 (@_unitedesque) reported 29 minutes ago

  Apple mail by birth, Gmail by notification issue.

• 
  Yava (@YavaCode) reported 29 minutes ago

  @Seti_Here so true. i also get so annoyed with that. also that "login pr register" while i just opened the app. like to SEE what you can give me, i need to open my gmail, click 10 links and see another slop SaaS ****

• 
  flanz 🍮 (@kamenvram) reported 30 minutes ago

  yo im so cooked i got hacked got my gmail deleted (a day ago) and i cant recover it becuase of googles ******* "google doesn't provide another way to sign in to this account" damn you youtube why do you still have to charge my memberships even after my whole yt channels gone.

• 
  superapt (@superapt2) reported 32 minutes ago

  Hi @teamyoutube, I'm having some trouble with my Gmail. Would you be able to assist?

• 
  FSD Life (@Tatakae360) reported 34 minutes ago

  Grok has a connector for Gmail and I use it for search. Also to find and organize invoices, things I may be overpaying for, ways to cut costs, etc. It’s infinitely better than the broken built in Gmail search

• 
  Don March (@marchland) reported 34 minutes ago

  I've been deleting emails in my Gmail account like crazy, and am down to 26,808. I'm making progress, but please help!

• 
  MissNovember❤️ (@Fulunem) reported 38 minutes ago

  Mme Johnson is in serious PAIN!! Someone call 10111!!! Tell them she just got shot down!!! Xem she won’t say a word on her Gmail shenanigans!!! What happens if a decision is taken to not prosecute her? The matter is dropped just like that? #MadlangaCommission #AndreaJohnson

• 
  Shep (@ShepRules) reported 41 minutes ago

  @AbakpaJob Just set up another Gmail account and tie your photos and Google drive to that one. Problem solved!