Skip to main content
by Meysam Azad
24 min read

Apple’s “User Not Found” Bounces: Check the Suppression List You Keep

Apple user not found bounces leave a trace: an address that received your mail before November 17 may sit on a suppression list under a reply that reads like a final answer, and it stays there until the entry expires or someone removes it. The recipient is what you lose in the meantime.

Remove an Apple entry only if it carries the user not found text, was created in the window your provider gives for when it suppressed Apple addresses (or from your first such reply on November 17 if it gave none), and the address delivered before; for Apple domains other than privaterelay.appleid.com, the Sign in with Apple relay, also find a second signal. Never remove unsubscribes or complaints. The eight steps are under the heading on checking and cleaning the list you keep.

What did Apple’s mail servers return on November 17?

An SMTP (RFC 5321) reply is the answer a receiving server gives a sending one.

SparkPost says that starting on the morning of November 17, Apple’s mail servers returned a large number of user-not-found bounces for iCloud and Hide My Email (Private Relay) addresses that are valid, and that this was an issue on Apple’s side.

The reply, as SparkPost prints it, is 550 5.1.1 <address>: user not found, bounce class 10. In its update of November 17, 20:40 UTC, SparkPost names privaterelay.appleid.com as its confirmed scope and says Apple returned the errors in several waves, with volume well below its peak but still occurring. SparkPost says valid recipients may have landed on suppression lists because the bounces looked like permanent failures.

Mailgun’s incident names iCloud and Private Relay addresses and calls the responses erroneous. Its page gives no domain list, no window, no cause and no SMTP reply code. We found no Apple statement about the bounces, and no page says which step of the SMTP conversation returned the reply.

Separately, AppleInsider reports that Apple’s status page listed iCloud Account & Sign In, iCloud Mail and iCloud.com entries that day. MacRumors reports that Apple told Bloomberg an internal update caused outages and that it was not a cyberattack. No source we found links that outage to the bounces, and it is not shown to be the cause.

Which Apple addresses were affected: privaterelay.appleid.com, icloud.com or private.icloud.com?

SparkPost confirmed privaterelay.appleid.com. Neither provider lists an affected domain besides it: Mailgun’s page and SparkPost’s first update say iCloud addresses without naming a domain. So check all of them, including me.com, mac.com, private.icloud.com and custom domains hosted on iCloud. The table lists the MX hosts each one uses.

Domain or familyWhat it isMXWhat the sources say about November 17
privaterelay.appleid.comSign in with Apple relay addresses. Apple said in August that new ones would move to private.icloud.com later in 2026 and that existing ones keep working; its support page now lists both domainssmtp1.privaterelay.appleid.com, smtp2.privaterelay.appleid.com, smtp3.privaterelay.appleid.com, smtp4.privaterelay.appleid.comSparkPost’s confirmed scope. The only domain in SparkPost’s incident of August 26 to August 27 too
private.icloud.comThe new relay domain. Apple’s June notice told email providers to add it to filtering, suppression lists and routing rules; Apple now marks that notice outdated, so read it as June guidancesmtp1.private.icloud.comNot reported either way. Apple’s support page says a Sign in with Apple Hide My Email address is created on this domain or on privaterelay.appleid.com
icloud.comiCloud mailboxes, and the addresses of iCloud+ Hide My Email, which stay on this domain. A Hide My Email alias cannot be told from a mailbox by its domainmx01.mail.icloud.com, mx02.mail.icloud.comMailgun’s page and SparkPost’s first update name iCloud addresses; SparkPost’s confirmed scope is the relay only. Unresolved. SparkPost’s August incident left it unaffected
me.com, mac.comAddresses or aliases of the same iCloud account, depending on when it was createdmx01.mail.icloud.com, mx02.mail.icloud.comAs icloud.com
A custom domain on iCloud MailA domain you own, pointed at iCloud Mailmx01.mail.icloud.com or mx02.mail.icloud.comNo provider page we read says custom domains were affected. Find them by MX
Apple recipient domains to check, with the MX hosts each one publishes.

“Hide My Email” names two features. The Sign in with Apple option creates addresses on privaterelay.appleid.com or private.icloud.com. The iCloud+ feature issues addresses on icloud.com; Apple’s August update says “iCloud+ Hide My Email addresses will remain on icloud.com”. iCloud Private Relay is a third thing: Apple describes it as part of iCloud+ that helps protect your privacy when you browse the web in Safari.

Our reading: the relay and the mailboxes use separate inbound hosts, and a custom domain on iCloud Mail shares the mailbox hosts. That is why a filter on Apple’s domain names can miss a custom domain. Neither point is a finding about the incident.

To find custom domains, look up each recipient domain’s MX with the MX lookup, or run a list of domains through the script below.

find-icloud-hosted-domains.sh bash
# Which of your recipient domains are hosted at iCloud Mail? Input is a file of
# distinct recipient DOMAINS, one per line (not addresses). A custom domain on
# iCloud Mail points its MX at the same two hosts as icloud.com, so a filter on
# Apple's own domain names misses it.
set -euo pipefail

DOMAINS_FILE="${1:?usage: find-icloud-hosted-domains.sh recipient-domains.txt}"
ICLOUD_MX='^[0-9]+ (mx01|mx02)\.mail\.icloud\.com\.$'

while IFS= read -r domain; do
  [ -n "$domain" ] || continue
  mx=$(dig +short MX "$domain" || true)
  if grep -qiE "$ICLOUD_MX" <<< "$mx"; then
    echo "$domain"
  fi
done < "$DOMAINS_FILE"

Why does your provider’s cleanup not reach your own suppression list?

SparkPost’s postmortem says its cleanup removed entries from SparkPost’s suppression list, and that if you also keep your own list, “you may need to remove affected addresses there too”. Bird, which mirrors SparkPost’s status posts, carries the same text. Which entries depends on how you build your list; the SparkPost section below gives the filters for a copy and for events.

Of the providers whose pages we read, only SparkPost, mirrored by Bird, says its cleanup covered only its own list. Mailgun’s page says “Impacted addresses have been removed from suppression lists.” and gives no guidance on lists a sender keeps.

Time limits make a copy drift further. AhaSend, the sending service DMARCguard uses, suppresses most bounces for 30 days, temporary failures for 7 and messages that expire after repeated attempts for 15. Each suppression carries an expires_at, and AhaSend’s webhook guide says it is worth storing so a UI can show when a suppression ends. AhaSend’s status page lists no incident for November 17: this is how one sending service expires entries, not a claim that AhaSend was affected.

Our arithmetic, conditional on AhaSend classing a 5.1.1 of November 17 that SparkPost calls wrong as an ordinary bounce: its own suppression would lapse on December 17, 2026. A mirrored copy in an application database would not lapse unless the application stored and honours expires_at. We store the expiry AhaSend sends and keep no second copy of a suppression that could outlive it.

How do you check and clean the suppression list you keep?

Pull the entries created from the start of your provider’s window, narrow them to Apple recipient domains and the 5.1.1 reply, leave every unsubscribe and complaint suppressed, and remove only entries that pass the removal test in step 5. Eight steps:

  1. Find every system that keeps suppressions. Your sending service, application database, CRM, data warehouse and any mail server you run: a provider’s cleanup does not reach them. Amazon SES keeps a global list and an account-level list. Its global list cannot be queried or edited, and an address drops off it on its own within 14 days; the account-level list is the one you can check.

  2. Pull the entries created from the start of the window. The window is the period your provider gives for when it added Apple addresses to suppression lists. SparkPost’s opens at 13:00 UTC on November 17. If your provider gave none, start from your first 5.1.1 to an Apple domain that day. A list you build from bounce events has no end time, because SparkPost says Apple was still returning some errors at 00:06 UTC on November 18.

  3. Narrow to Apple recipient domains. Select recipients at icloud.com, me.com, mac.com, privaterelay.appleid.com or private.icloud.com, or at any domain whose MX host is mx01.mail.icloud.com or mx02.mail.icloud.com.

  4. Match the reply. Narrow to entries that carry 550 5.1.1 and the text user not found, which leaves out the relay’s other 5.1.1 replies.

  5. Apply the removal test. Unsubscribes, complaints and other bounce reasons stay suppressed. Google and Yahoo’s bulk-sender rules require one-click unsubscribe for bulk marketing mail, and Apple’s postmaster page tells bulk senders not to reactivate addresses on an unsubscribe or suppression list. Our reading: that is about addresses someone chose to leave, while SparkPost and Mailgun describe these entries as created by wrong replies.

    The test is our advice, not a source’s. For privaterelay.appleid.com, the only domain a provider confirmed, an entry qualifies when it was created in the window, carries the user not found text, and the address delivered before the window.

    For every other Apple domain, including custom domains on iCloud MX, that is not enough. No provider names them, and a mailbox deleted that week can pass the test too. Add a second signal from your own data: a cluster of addresses at that domain, each of which had delivered recently, bouncing with the same reply within the same hours, or the recipient telling you mail stopped. That is a judgement call, not a threshold.

  6. Remove only the entries that pass the test. The test cannot prove a reply was wrong: an address deleted that week can pass it too. If an entry whose address is in fact gone is removed, the next send is one more hard bounce to Apple. On Mailgun, the documentation says delivery to a removed address resumes until it bounces again; what happens elsewhere depends on the provider, and SparkPost says it no longer adds addresses for these Apple errors. Remove one recipient and one sending domain at a time: on AhaSend always pass --domain, and on Mailgun delete the single address.

  7. Stop it coming back. Sync jobs, webhook replays and imports can restore entries. Route a sudden spike of 5.1.1 replies to one mailbox provider to review before it becomes a suppression; our post on hard-bounce thresholds and bounce monitoring covers the monitoring side.

  8. Reissue time-bound mail. Send a new one-time code or reset link instead of replaying a message that bounced on November 17.

We ran the scripts below on made-up data shaped like each vendor’s documented response, and the SQL on a throwaway database. None has run against a live account, and the AhaSend delete command has not been run, so start with the counts. Each script prints counts only. Where a script writes entries to a file, only you can read it: delete the file when you finish, and keep its contents out of tickets, chats and AI tools.

SparkPost

SparkPost says it removed these entries from its own list, so the live list no longer holds them; the filters below are for a copy or export you hold, or your event feed. For a copy: source Bounce Rule, a recipient at privaterelay.appleid.com, created in the window, ending 21:05 UTC for US accounts and 21:16 UTC for EU accounts. For events: bounce class 10 and the reply, from 13:00 UTC, with no end.

The script reads a JSON export in the shape of SparkPost’s Suppression List search response and calls nothing. That API treats from and to as the date an entry was last updated, not created, so the script applies the window to each entry’s created field. It counts entries at all five Apple domains so you can see them; SparkPost’s postmortem names only the relay.

sparkpost-apple-suppressions.sh bash
# SparkPost: the entries in YOUR copy of the suppression list that fall inside
# SparkPost's window. The input is a JSON file you already hold: an export or a
# copy of the Suppression List search response, whose result entries carry
# `recipient`, `source` and `created`. SparkPost says it removed these entries
# from its own list, so the live list is not where to look. No call is made to
# SparkPost. Prints counts only.
#
# The window applies to each entry's `created` field. EU accounts use the EU
# end time below.
set -euo pipefail

INPUT="${1:?usage: sparkpost-apple-suppressions.sh export.json}"
FROM="2026-11-17T13:00:00Z"
TO="2026-11-17T21:05:00Z" # 2026-11-17T21:16:00Z for EU accounts
SOURCE="Bounce Rule"
APPLE_DOMAINS="icloud.com me.com mac.com privaterelay.appleid.com private.icloud.com"

jq -r --arg from "$FROM" --arg to "$TO" --arg source "$SOURCE" --arg domains "$APPLE_DOMAINS" '
  ($domains | split(" ")) as $apple
  | (if type == "array" then . else .results end)[]
  | select(.source == $source)
  | (.recipient | ascii_downcase | split("@")[1]) as $recipient_domain
  | select($recipient_domain | IN($apple[]))
  | (.created | sub("\\.[0-9]+"; "") | sub("\\+00:00$"; "Z") | fromdateiso8601) as $created
  | select($created >= ($from | fromdateiso8601) and $created <= ($to | fromdateiso8601))
  | $recipient_domain' "$INPUT" | sort | uniq -c | sort -rn

Mailgun

Mailgun’s page gives no window, so start from your own first 5.1.1 to an Apple domain on November 17; 20:31 UTC is only when Mailgun posted. The bounces API lists a domain’s bounces and has no recipient-domain filter, so the script filters in place. Deleting one address, DELETE /v3/{domain_name}/bounces/{address}, resumes delivery to it until it bounces again.

mailgun-apple-bounces.sh bash
# Mailgun: bounces of one sending domain, selected when the recipient is at an Apple
# domain and the bounce was created from FROM (UTC). Mailgun's page gives no
# window, so set FROM to your own first 5.1.1 to an Apple domain on November 17.
# Prints counts. The entries stay in a file that only you can read: delete it
# when the check is done, and keep its contents out of tickets, chats and AI tools.
#
# The list endpoint has no recipient-domain filter (`term` matches the start of
# an address), so the Apple filter runs here. EU accounts use
# https://api.eu.mailgun.net/v3. Items carry address, code, error and created_at
# (RFC 822); `paging.next` is the URL of the next page.
set -euo pipefail
umask 077

API="https://api.mailgun.net/v3"
DOMAIN="${MAILGUN_DOMAIN:?set MAILGUN_DOMAIN to the sending domain}"
: "${MAILGUN_API_KEY:?set MAILGUN_API_KEY}"
FROM="${FROM:?set FROM to the UTC time of your first 5.1.1, as YYYY-MM-DDTHH:MM:SSZ}"
APPLE_DOMAINS="icloud.com me.com mac.com privaterelay.appleid.com private.icloud.com"
OUT="mailgun-apple-bounces.ndjson"
: > "$OUT"

url="$API/$DOMAIN/bounces?limit=1000"
while [ -n "$url" ]; do
  body=$(curl -sS --fail --user "api:$MAILGUN_API_KEY" "$url")
  [ "$(printf '%s' "$body" | jq '.items | length')" -gt 0 ] || break
  printf '%s' "$body" | jq -c --arg from "$FROM" --arg domains "$APPLE_DOMAINS" '
    ($domains | split(" ")) as $apple
    | .items[]
    | (.address | ascii_downcase | split("@")[1]) as $recipient_domain
    | select($recipient_domain | IN($apple[]))
    | select((.created_at | strptime("%a, %d %b %Y %H:%M:%S %Z") | mktime)
             >= ($from | fromdateiso8601))
    | . + {recipient_domain: $recipient_domain}' >> "$OUT"
  next=$(printf '%s' "$body" | jq -r '.paging.next // empty')
  [ "$next" != "$url" ] || break
  url="$next"
done

echo "bounces selected: $(wc -l < "$OUT")"
echo
echo "by recipient domain, and whether the error has the 5.1.1 text:"
jq -r '[.recipient_domain,
        (.error | ascii_downcase | (contains("5.1.1") and contains("user not found")))]
       | @tsv' "$OUT" | sort | uniq -c | sort -rn

AhaSend, the service we send through

The CLI lists suppressions created from a time, 100 per page, and the response carries pagination.next_cursor. Its --domain flag is the sending domain, not the recipient’s, so the script filters the recipient domain itself and skips entries AhaSend marks protected: unsubscribes and complaints. Needs the suppressions:read scope.

ahasend-list-apple-suppressions.sh bash
# AhaSend: suppressions created since the window opened, selected only when the
# recipient is at an Apple domain and AhaSend does not mark the entry protected.
# Prints counts. The entries stay in a file that only you can read: delete it
# when the check is done, and keep its contents out of tickets, chats and AI tools.
#
# `--domain` on `list` is the SENDING domain, not the recipient's, so the Apple
# filter runs here. Needs an API key with the suppressions:read scope.
set -euo pipefail
umask 077

# SparkPost's window start; use your first 5.1.1 to an Apple domain if you have no window.
FROM="2026-11-17T13:00:00Z"
APPLE_DOMAINS="icloud.com me.com mac.com privaterelay.appleid.com private.icloud.com"
OUT="apple-suppressions.ndjson"
: > "$OUT"

cursor=""
while :; do
  args=(suppressions list --from-time "$FROM" --limit 100 --output json)
  [ -n "$cursor" ] && args+=(--after "$cursor")
  page=$(ahasend "${args[@]}")
  printf '%s' "$page" | jq -c --arg domains "$APPLE_DOMAINS" '
    ($domains | split(" ")) as $apple
    | .data[]
    | (.email | ascii_downcase | split("@")[1]) as $recipient_domain
    | select($recipient_domain | IN($apple[]))
    | select(.protected == false)
    | . + {recipient_domain: $recipient_domain}' >> "$OUT"
  [ "$(printf '%s' "$page" | jq -r '.pagination.has_more')" = "true" ] || break
  cursor=$(printf '%s' "$page" | jq -r '.pagination.next_cursor')
done

echo "entries selected: $(wc -l < "$OUT")"
echo
echo "by recipient domain, sending domain, and whether the reason has the 5.1.1 text:"
jq -r '[.recipient_domain, .domain,
        (.reason | ascii_downcase | (contains("5.1.1") and contains("user not found")))]
       | @tsv' "$OUT" | sort | uniq -c | sort -rn

To remove one entry that passes the test, run ahasend suppressions delete recipient@example.net --domain send.example.com, with the recipient’s address and your own sending domain. Needs the suppressions:delete scope. The CLI asks you to confirm each removal. Pass --domain every time: the flag is optional, and without it the command deletes the address’s suppressions on every domain, unsubscribes and complaints included.

Your own database or CRM

Count first, by recipient domain and sending domain, then remove by the ids you reviewed. The first query prints no address and reduces the reason text to a yes or no.

count-apple-suppressions.sql sql
-- Suppressions at Apple recipient domains, created from the start of the window,
-- counted by recipient domain and sending domain. Protected entries
-- (unsubscribes, complaints) are left out. Counts only: no address is printed,
-- and the reason text can hold one, so it is reduced to a yes or no.
--
-- The window start is SparkPost's. If your provider gave none, use the time of
-- your first 5.1.1 to an Apple domain on November 17.
--
-- Assumed table: suppressions(id, email, sending_domain, reason, source,
-- created_at, protected). created_at holds ISO 8601 UTC. The SQL is written for
-- PostgreSQL and SQLite; adapt the names to your own schema.
WITH apple(domain) AS (
  VALUES ('icloud.com'), ('me.com'), ('mac.com'),
         ('privaterelay.appleid.com'), ('private.icloud.com')
)
SELECT
  apple.domain AS recipient_domain,
  s.sending_domain,
  CASE
    WHEN lower(s.reason) LIKE '%5.1.1%' AND lower(s.reason) LIKE '%user not found%'
      THEN 'yes'
    ELSE 'no'
  END AS has_5_1_1_user_not_found,
  COUNT(*) AS entries
FROM suppressions AS s
JOIN apple ON lower(s.email) LIKE '%@' || apple.domain
WHERE s.created_at >= '2026-11-17T13:00:00Z'
  AND NOT s.protected
GROUP BY apple.domain, s.sending_domain, has_5_1_1_user_not_found
ORDER BY entries DESC;

The second ends in a rollback: run it once, compare the count with the ids you reviewed, then switch it to a commit.

remove-reviewed-suppressions.sql sql
-- Remove only the entries that pass the removal test, by id. The ids come from
-- your own review. For privaterelay.appleid.com, the only domain a provider
-- confirmed, an entry qualifies when it was created in the window, carries the
-- user not found text, and the address delivered before the window. For every
-- other Apple domain that test is not enough: no provider names these domains,
-- and a mailbox deleted that week can pass it too. Add a second signal from your
-- own data first: a cluster of addresses at that domain, each of which had
-- delivered recently, bouncing with the same reply within the same hours, or the
-- recipient telling you mail stopped. That is a judgement call, not a
-- threshold. The test cannot prove a reply was wrong. If an entry whose address
-- is in fact gone is removed,
-- the next send is one more hard bounce to Apple. On Mailgun, the documentation
-- says delivery to a removed address resumes until it bounces again; what
-- happens elsewhere depends on the provider, and SparkPost says it no longer
-- adds addresses for these Apple errors. The guards below keep a wrong id from
-- removing an unsubscribe, a complaint or a non-Apple address.
--
-- The window start is SparkPost's. If your provider gave none, use the time of
-- your first 5.1.1 to an Apple domain on November 17. Same assumed table as
-- count-apple-suppressions.sql. The script ends in ROLLBACK: run it once,
-- compare the first count with the number of ids you reviewed, then change the
-- last line to COMMIT.
WITH apple(domain) AS (
  VALUES ('icloud.com'), ('me.com'), ('mac.com'),
         ('privaterelay.appleid.com'), ('private.icloud.com')
)
SELECT COUNT(*) AS would_remove
FROM suppressions AS s
WHERE s.id IN (-1, -2, -3) -- replace with the ids you confirmed
  AND NOT s.protected
  AND s.created_at >= '2026-11-17T13:00:00Z'
  AND lower(s.reason) LIKE '%5.1.1%'
  AND lower(s.reason) LIKE '%user not found%'
  AND EXISTS (SELECT 1 FROM apple WHERE lower(s.email) LIKE '%@' || apple.domain);

BEGIN;

WITH apple(domain) AS (
  VALUES ('icloud.com'), ('me.com'), ('mac.com'),
         ('privaterelay.appleid.com'), ('private.icloud.com')
)
DELETE FROM suppressions
WHERE id IN (-1, -2, -3) -- the same ids
  AND NOT protected
  AND created_at >= '2026-11-17T13:00:00Z'
  AND lower(reason) LIKE '%5.1.1%'
  AND lower(reason) LIKE '%user not found%'
  AND EXISTS (SELECT 1 FROM apple WHERE lower(suppressions.email) LIKE '%@' || apple.domain);

ROLLBACK; -- COMMIT once the count above matches your reviewed ids

What should you not do?

  • Do not unsuppress every Apple address. A relay 5.1.1 can have an ordinary cause, such as the user turning forwarding off.
  • Do not clear all bounces or wipe a list. Mailgun’s clear-all call and AhaSend’s wipe command are not part of this check.
  • Do not run an AhaSend delete without --domain.
  • Do not replay time-bound mail.
  • Do not paste recipient addresses into tickets, chats or AI tools. The count queries above need none.

Why is a 550 5.1.1 reply suppressed as a hard bounce?

In 550 5.1.1, the second part is the enhanced status code. RFC 3463 §3.2 defines X.1.1 as a bad destination mailbox address and says: “The mailbox specified in the address does not exist.” It adds that the code suits permanent failures only.

The first part is the reply code. Under RFC 5321 §4.2.1 a 5yz reply is a permanent negative completion, and the client SHOULD NOT repeat the exact request. The same section notes: “Even some “permanent” error conditions can be corrected”.

Sending services act on that. Mailgun’s documentation says its bounce list holds permanent recipient mailbox errors such as a non-existent mailbox, and AhaSend’s defines a hard bounce the same way and suppresses most bounces for 30 days. The cost is that a mistaken one removes a real recipient.

A 5.1.1 from the Apple relay can have an ordinary cause. Apple’s developer support says a 550 5.1.1 unknown user reply likely means the user turned off forwarding for that app, and Resend lists a deleted Hide My Email address and a daily limit among the usual reasons a relay address bounces. The relay also answers 550 5.1.1 with other text: Apple’s developer support ties unauthorized sender to sender registration and authentication, which is why step 4 matches the user not found text. Our reading: a relay 5.1.1 is not wrong by default, which is why a blanket unsuppress is wrong.

What is the timeline of the Apple bounces (UTC)?

SparkPost’s suppression window opens at 13:00 UTC on November 17. That is the start of SparkPost’s window, not of Apple’s errors: Mailgun gave none. SparkPost’s postmortem, posted at 00:06 UTC on November 18, says Apple was still returning some of these errors intermittently, and we found no later statement from SparkPost or Mailgun saying they stopped.

The providers post with different offsets: SparkPost -05:00, Bird +01:00 and Mailgun -08:00. Every time below is converted to UTC.

WhenWhat happenedSource
November 17, 13:00 UTCStart of the window in which SparkPost added Apple relay addresses to its suppression list. SparkPost’s window, not Apple’sSparkPost
November 17, 19:45 UTCSparkPost’s first update: valid iCloud and Hide My Email addresses; volume had dropped sharply over the past few hoursSparkPost
November 17, 20:31 UTCMailgun posts that it is removing addresses added to suppression lists during the affected timeMailgun
November 17, 20:40 UTCSparkPost: Apple-side, addresses valid, scope privaterelay.appleid.com, cleanup startedSparkPost
21:05 UTC (US), 21:16 UTC (EU)SparkPost stops adding these addresses to its list. Not the end of the incidentSparkPost
November 17, 22:55 UTCSparkPost: cleanup finished, and a change so that if Apple returns these errors again the addresses are not added. Bird posts the same updates at 22:09 UTC and 22:54 UTCSparkPost, Bird
November 18, 00:06 UTCPostmortem: guidance for senders who keep their own list; Apple still returning some errors intermittentlySparkPost
November 18, 03:58 UTCMailgun: addresses are being removed from suppression listsMailgun
November 18, 19:35 UTCMailgun: impacted addresses removed from suppression listsMailgun
Provider updates on the November 17 bounces. All times UTC.

Apple’s status page the same day is shown separately, because it is not part of the providers’ account:

WhenWhat the status page showedSource
November 17, 16:45 UTCiCloud Account & Sign In listed as an issue. An archived copy of the feed from 17:50 UTC shows it, and shows no iCloud Mail entryArchived feed
November 17, 19:03 UTCiCloud Mail and iCloud.com added, for some users. Secondary sourceAppleInsider
Apple's status page on November 17, shown separately from the provider updates. All times UTC.

No source we found links Apple’s outage to the bounces. The 16:45 UTC row rests on the archived copy and the 19:03 UTC row on AppleInsider’s report.

Were there earlier Apple incidents in 2026?

This is the third Apple incident this year that we found on SparkPost’s status page, with a suppression cleanup each time. They differ in what the pages say.

  • April 5: SparkPost attributed increased bounces and delays to Apple domains to an Apple iCloud Mail outage, then removed the Apple and Apple Private Relay suppressions from Apple’s incident window. Its cleanup finished at 01:00 UTC on April 6. The page names no SMTP reply.
  • August 26 to August 27: SparkPost saw 550 5.1.1 user not found from the relay, privaterelay.appleid.com, from around 23:00 UTC on August 26. It says icloud.com, me.com and mac.com were unaffected, that a substantial share of the addresses had delivered successfully as recently as August 26, and that it would remove the suppressions itself. It resolved the incident at 22:56 UTC on August 27.
  • November 17: this incident. Only this incident and August’s name a 5.1.1 reply.

SparkPost announced the change, so that these Apple errors no longer add addresses to its list, in the update that closed this cleanup.

Why can’t your DMARC reports show this?

Apple’s postmaster page for iCloud Mail offers no feedback loop and does not mention aggregate reports. Our own guide to aggregate reports lists Apple/iCloud as not a documented report sender, and our own intake agrees.

Key finding

0 DMARC aggregate reports from Apple have reached DMARCguard for its own domains.

Source: DMARCguard report intake

That absence means something because we do send to iCloud addresses: we sent mail to icloud.com addresses before November 17. Reports from other mailbox providers kept arriving through the days of the incident. If reports from other mailbox providers have stopped for you, check whether your reports are reaching you.

From 13:00 UTC on November 17 to the end of our data, 01:26 UTC on November 21, we sent no mail to an icloud.com, me.com, mac.com, privaterelay.appleid.com or private.icloud.com address, so we received no bounce and no suppression event for one. The check covered those five domains by name, not custom domains hosted on iCloud Mail’s MX hosts. No Apple entry was added to our own suppression list, because we had nothing in flight to Apple in the window, not because our setup was immune. We cannot say what Apple sent or did not send to anyone else.

What this supports: DMARC monitoring cannot show this incident. Your bounce events and suppression list are the record. SPF (RFC 7208), DKIM (RFC 6376) and DMARC (RFC 9989) concern who sent a message; a user-not-found reply concerns the recipient’s mailbox. Nothing on SparkPost’s or Mailgun’s pages ties the bounces to sender authentication.

How is this different from the two Microsoft incidents?

Three mailbox-provider incidents between November 6 and November 17 produced three different SMTP answers and three different jobs for a sender.

RowAS9031519 (EX1545193)EX1550543November 17 (Apple)Basis
DatesMicrosoft’s: November 6, 18:00 UTC to November 9, 11:02 UTCMicrosoft’s final posts: November 12, 10:05 UTC to November 13, 16:10 UTCSparkPost’s window from November 17, 13:00 UTC; no end statedSourced
Reply550 5.7.1 or 550 5.7.23, per Microsoft451 4.7.0, 421 4.3.2, 451 4.7.700, SparkPost’s list, which may not be complete550 5.1.1 user not foundSourced
ClassPermanent, policy (X.7)TemporaryPermanent, address (X.1)Our reading
What a sender’s system didBounced; may suppressQueued and retriedBounced; may suppress the recipientOur reading
What to doFix DMARC, resend what bouncedResend only what expiredRemove the entries that pass the removal test; resend only what still mattersOur reading
What persistsThe messageThe message, if it expiredThe recipient, until the entry expires or is removedOur reading
Three November 2026 mailbox-provider incidents side by side. Rows marked our reading are ours, not a source's.

The AS9031519 bounces and our post on EX1550543 cover the Microsoft cases; the second lists the same suppression check as its step 5. The table has no cause row for Apple: we found no Apple explanation of the bounces.

A fourth event followed within days: Yahoo country domains moved to a new MX host, and Yahoo said senders may see a spike in error responses such as a temporary 450. That one calls for waiting, not for a list check: which Yahoo domains moved and what a 450 means.

Frequently asked questions

What does 550 5.1.1 user not found mean?

A 550 5.1.1 reply is a permanent refusal: RFC 3463 §3.2 defines X.1.1 as a bad destination mailbox address, and RFC 5321 §4.2.1 says the sender should not repeat the exact request. Sending services treat it as a hard bounce and suppress the address. On November 17, 2026, SparkPost said Apple returned it for valid recipients.

Did Apple confirm what caused the bounces?

No. We found no Apple statement about the bounces. MacRumors reports that Apple told Bloomberg an internal update caused outages the same day and that it was not a cyberattack. No source we found links that outage to the bounces.

Which Apple addresses were affected?

SparkPost’s confirmed scope is privaterelay.appleid.com, the Sign in with Apple relay. Neither provider lists another affected domain: Mailgun’s page and SparkPost’s first update say iCloud addresses without naming a domain. Check icloud.com, me.com, mac.com, private.icloud.com and any domain whose MX is mx01 or mx02.mail.icloud.com.

Is Apple still returning the errors?

SparkPost’s last statement, its postmortem posted at 00:06 UTC on November 18, 2026, says Apple was still returning some of these errors intermittently. We found no later statement from SparkPost or Mailgun saying they stopped.

Will my provider’s cleanup fix my own suppression list?

Not by itself. SparkPost says its cleanup covered only its own list and that senders who keep their own may need to remove the affected addresses there too. Mailgun’s page says its addresses were removed from suppression lists and does not mention lists a sender keeps.

Should I unsuppress every iCloud address?

No. For privaterelay.appleid.com, the domain SparkPost confirmed, remove only entries created in your provider’s window with the user not found text whose address delivered before. Other Apple domains need a second signal, because no provider names them. Apple’s developer support says a 550 5.1.1 unknown user reply from the relay likely means forwarding was turned off. Unsubscribes and complaints stay suppressed.

What are privaterelay.appleid.com and private.icloud.com?

privaterelay.appleid.com is the relay domain for Sign in with Apple addresses. Apple said in August 2026 that new relay addresses would be issued on private.icloud.com starting later in 2026 and that existing ones keep working. Apple’s June notice asked email providers to add private.icloud.com to filtering and suppression rules; Apple now marks it outdated. Apple’s support page for Sign in with Apple says a Hide My Email address is created on private.icloud.com or privaterelay.appleid.com.

Would DMARC reports have shown this?

No. DMARC aggregate reports carry authentication results, not a recipient server’s replies, and Apple’s postmaster page documents no aggregate reporting. None of the aggregate reports that reach DMARCguard for its own domains comes from Apple. The bounces appear in your bounce events and suppression list. Nothing on SparkPost’s or Mailgun’s pages links authentication to them.

What should you do about the Apple bounces?

  • A provider’s cleanup is not your list’s cleanup: of the providers whose pages we read, only SparkPost says so. Check the lists you keep.
  • Narrow the Apple recipients created in your provider’s window to the user not found reply, and remove an entry only if it passes the test in step 5. Leave unsubscribes and complaints suppressed, and never wipe a list.
  • DMARC aggregate reports will not show Apple user not found bounces; your bounce events and suppression lists are the record.

Nothing in this incident calls for a DNS change. To see each DMARC report as it arrives, and each gap, start monitoring your DMARC reports: free plan, no credit card.

Updates

  • November 21, 2026. First published. SparkPost’s latest statement on the incident is its postmortem of 00:06 UTC on November 18; Mailgun’s is its resolution at 19:35 UTC on November 18; we found no Apple statement about the bounces.