Opens in a new tab

DMARCbis is here: what RFC 9989 changes in your DMARC record, and how to update it

14 min read

In May 2026 the IETF published the revision of DMARC that the email world has been calling DMARCbis for years. It replaces RFC 7489, the 2015 document almost every DMARC record on the internet was written against, and it comes in three parts:

  • RFC 9989: DMARC itself. The record, the policy, and how receivers find your record. It obsoletes RFC 7489 and RFC 9091.
  • RFC 9990: aggregate reports, the daily XML files your rua= address receives.
  • RFC 9991: failure reports, the per-message reports your ruf= address may receive.

All three are on the IETF Standards Track. RFC 7489 was only ever an Informational document: DMARC was everywhere, but it was not formally a standard. Now it is.

The good news first: a DMARC record that works today almost certainly keeps working. Nobody is going to reject your mail because your record was written for RFC 7489. But three tags you may be using are gone, three new ones exist, the way receivers find the record for a subdomain has changed, and a few common mistakes now cost more than they used to. This guide goes through each change, what it means for your record, and exactly what to do about it.

Check your own record while you read: the DMARC checker reads it the way RFC 9989 tells receivers to. It names any tag that is no longer part of DMARC, shows which record a subdomain actually inherits, and checks that your report addresses are authorised.

The short version

  • Remove pct=, rf= and ri=. They are no longer part of DMARC, and receivers that follow RFC 9989 ignore them.
  • If you used pct for a gradual rollout, use t=y instead. It is test mode: receivers apply one level below your published policy.
  • Consider np=reject. It is a policy for subdomains that do not exist, which are almost always forged.
  • Proofread p=, sp= and np=. A typo in any of them now makes receivers treat your whole record as p=none.
  • Write v=DMARC1 exactly like that, first. The version is case-sensitive.
  • Remove size limits such as !10m from your report addresses. That syntax is gone.
  • If you have DMARC records on subdomains, check which record each subdomain now gets. The rules for finding it changed.
  • No change needed: adkim, aspf, fo, rua, ruf and sp work as before. Most domains never need the new psd tag.

Why DMARC was revised

Ten years of running DMARC at scale showed where RFC 7489 did not match reality. The pct tag was applied inconsistently. Receivers found a domain’s “organisational domain” with the Public Suffix List, and different copies of that list gave different answers. Public suffix domains, such as registries, had no way to publish a policy for the names under them without an experimental extension (RFC 9091). And a single document covered the protocol and both report formats, so none of them could be updated without reopening all of them.

DMARCbis addresses each of those. The protocol itself did not change in spirit: you still publish a TXT record at _dmarc.yourdomain, mail still passes DMARC when SPF or DKIM passes and aligns with the From domain, and your policy still asks receivers what to do with mail that fails. SPF (RFC 7208) and DKIM (RFC 6376) are not affected.

What changed in the record, tag by tag

TagUnder RFC 9989What to do
vUnchanged. It must come first, and its value DMARC1 is case-sensitive.Write exactly v=DMARC1.
pNow “recommended” rather than required. When it is missing, a record with a working report address is read as p=none. Without one, receivers apply no DMARC at all.Always write p=.
spUnchanged: the policy for existing subdomains.Nothing.
npNew in the core standard (it comes from RFC 9091): the policy for subdomains that do not exist.Consider np=reject.
tNew: test mode, y or n (default n).Use it instead of pct for trials.
psdNew: whether the domain is a public suffix, y, n or u (default u).Leave it out unless you run a public suffix or a separate organisational boundary.
adkim, aspfUnchanged: r (relaxed, the default) or s (strict).Nothing.
foUnchanged: 0, 1, d, s. Ignored unless ruf is set.Nothing.
rua, rufUnchanged, except that the size limit suffix (!10m) was removed.Remove any ! size limits.
pctRemoved.Remove it, or replace it with t=y.
rfRemoved. Failure reports have one format.Remove it.
riRemoved. The report interval is the receiver’s to choose, and most send daily.Remove it.

pct is gone, and t=y replaces it

pct asked receivers to apply your policy to only a percentage of failing mail, so you could move from p=none to p=quarantine at 10%, then 50%, then 100%. Appendix A.6 of RFC 9989 explains why that did not survive. In practice only two values were applied reliably: 100, the default, and 0. Anything in between was handled differently from one receiver to the next. And pct=0 had quietly taken on a meaning of its own: mailing lists and other intermediaries treated it as a signal to rewrite the From address, and domain owners used that to measure how much of their mail passed through such intermediaries before they enforced.

So the revision kept the one useful behaviour and gave it a clear name: the t tag, for testing. RFC 9989 describes t=y and t=n as the counterparts of pct=0 and pct=100.

What happens to a record that still says pct

RFC 9989 no longer registers pct, and receivers must ignore tags that are not registered. A receiver that follows the new standard therefore applies your policy to all failing mail, whatever pct says, including pct=0. Nothing maps an old pct=0 onto t=y for you. Receivers that have not updated yet may keep honouring pct for a while, so a record with pct=25 is enforced in full by some receivers and sampled by others. That is the opposite of predictable, which is the whole point of a policy.

  • pct=100: remove it. It never changed anything.
  • pct between 1 and 99: decide what you meant. If your own mail passes, remove pct and let the full policy apply. If you are still trialling the policy, replace pct with t=y.
  • pct=0: you were testing. Write t=y instead.

How test mode works

t=y tells receivers you are trialling your policy, and asks them to apply one level below it:

You publishReceivers apply to failing mail
p=reject; t=yquarantine: the spam folder instead of a refusal
p=quarantine; t=ynone: delivered as usual, and reported
p=none; t=yno effect: t does nothing under p=none

Test mode does not change reporting: aggregate reports keep arriving, and under RFC 9990 they include a testing field, so you can see that receivers saw it. It also invites intermediaries such as mailing lists to apply their special handling, typically rewriting the From address. That is useful: your reports start to show how much of your mail travels through lists and forwarders before you enforce. You can read more on the t=y test mode page.

The new rollout, step by step

Where pct used to ramp from 10% to 100%, each step up is now published in test mode first, then made real:

  1. Monitor. v=DMARC1; p=none; rua=mailto:dmarc@example.com. Collect at least two weeks of reports and fix every legitimate sender that fails.
  2. Trial quarantine. v=DMARC1; p=quarantine; t=y; rua=mailto:dmarc@example.com. Failing mail is still delivered, and lists start rewriting.
  3. Quarantine. Remove t=y. Failing mail goes to spam.
  4. Trial reject. v=DMARC1; p=reject; t=y; rua=mailto:dmarc@example.com. Failing mail is quarantined, exactly as in step 3.
  5. Reject. Remove t=y. Failing mail is refused, and your domain is fully protected.

The most common mistake with test mode is forgetting to remove it. A record that says p=reject; t=y is quarantining, not rejecting, however long it stays that way.

np: a policy for subdomains that do not exist

Forgers love subdomains nobody created: billing.example.com, secure-login.example.com. The np tag, which started life in the experimental RFC 9091 and is now part of the core standard, lets you set a policy just for those: subdomains that do not exist in DNS at all.

When np is missing, receivers use sp; when sp is missing too, they use p. Like every DMARC policy, np only affects mail that fails DMARC. Legitimate mail from a name with no DNS records is extremely rare, and many receivers already refuse a From domain that does not resolve, so np=reject is usually safe even while your main policy is still p=none:

v=DMARC1; p=none; np=reject; rua=mailto:dmarc@example.com

That record monitors your real mail and refuses forgeries from invented subdomains. See sp= and np= for how the subdomain policies combine.

How receivers find your record: the DNS tree walk

This is the biggest change under the hood, and the one most likely to surprise anyone with DMARC records on subdomains.

Under RFC 7489, when mail came from news.example.com and there was no record at _dmarc.news.example.com, receivers looked up the “organisational domain” in the Public Suffix List. That is the list that says .com and .co.uk are suffixes, so example.com is the registered domain. Receivers used different copies of that list, updated at different times, and occasionally reached different answers.

RFC 9989 replaces the list with a DNS tree walk (section 4.10):

  1. The receiver looks up _dmarc. plus the From domain. If there is exactly one DMARC record there, that is your record.
  2. If not, it walks up the name one label at a time, looking up _dmarc. at each parent until it reaches the top-level domain. A name with eight or more labels jumps straight to its last seven, so a walk never takes more than eight lookups.
  3. A single record with psd=y or psd=n ends the walk early.
  4. The organisational domain is the name whose record says psd=n. Failing that, it is the name one label below a record that says psd=y. Failing that, it is the shortest name that has a record.
  5. The subdomain gets the organisational domain’s record, with sp or np applying. If the organisational domain has none, it gets the public suffix domain’s record.

For most domains the answer is the same as before. The difference shows when there are records at more than one level:

Mail fromRecords published atRecord applied under RFC 9989
a.mail.example.commail.example.com and example.comexample.com‘s, the shortest name with a record, not the nearer mail.example.com
a.mail.example.commail.example.com (with psd=n) and example.commail.example.com‘s, because psd=n marks it as its own organisational domain
news.example.co.ukexample.co.ukexample.co.uk‘s, the same as before
shop.example.comtwo records at shop.example.com, one at example.comexample.com‘s. Two records at one name count as none.

The first row surprises people most. If a part of your organisation runs mail.example.com as its own domain, with its own policy for everything beneath it, publish psd=n in its record. Otherwise its subdomains inherit the main domain’s record.

The organisational domain matters for alignment too. Under relaxed alignment (adkim=r, aspf=r), a DKIM signature or SPF domain aligns when it has the same organisational domain as the From address, and that is now decided by the same walk. See DMARC alignment for how relaxed and strict alignment differ.

Do you need the psd tag?

Almost certainly not. psd=y is for operators of public suffixes, such as registries, publishing a policy for the domains registered under them. psd=n is for a domain that should be treated as its own organisational domain, like the mail.example.com example above. Everyone else leaves it out, which means psd=u: let the walk decide.

Mistakes that now behave differently

RFC 9989 is precise about what receivers do with a record that is not quite right, and the answers are not all the same:

The mistakeWhat receivers do
A typo in p, sp or np, such as sp=rejctWith a working rua= address, they read the whole record as p=none. Without one, they apply no DMARC at all (section 4.10.1).
p= missingThe same: p=none if rua= works, otherwise no DMARC.
A bad value anywhere else, such as adkim=x, t=maybe or fo=zThey ignore that value and use the tag’s default (section 4.8). The rest of the record still works.
A report address that is not a valid URIThat address is ignored. The others still receive reports.
v=dmarc1, or v= not firstNot recognised as a DMARC record. It is ignored.
The same tag twice, or two DMARC records at one nameIgnored. A subdomain then falls back to its organisational domain’s record.
An unknown tag, including pct, rf and riIgnored. It does no harm, but it no longer does anything.

The first row is the one to worry about. v=DMARC1; p=reject; sp=rejct; rua=mailto:dmarc@example.com looks like a reject policy, but receivers following RFC 9989 read it as p=none for the whole domain. A single typo in a subdomain tag switches off the protection you thought you had.

Reports: what RFC 9990 and RFC 9991 change

Aggregate reports (rua)

Aggregate reports are still compressed XML files, one per receiver per reporting period, usually daily. RFC 9990 tightens the format and adds a few fields worth knowing about:

  • testing: whether your record had t=y when it was evaluated.
  • np: your non-existent subdomain policy.
  • discovery_method: treewalk or psl, which tells you whether that receiver already finds records the RFC 9989 way.
  • The DKIM selector is now required for every signature reported, which makes it much easier to tell which of your services signed a message.
  • A reason code, policy_test_mode, for mail that was spared because of t=y.

One rule is unchanged, and it catches people out constantly. If your rua= address is at another organisation’s domain, such as a report service, that domain has to authorise it. It does so by publishing yourdomain._report._dmarc.theirdomain as a TXT record containing v=DMARC1. Without it, receivers silently drop your reports. The name starts with the domain your record was found at, so for a subdomain that inherits its record, that is the organisational domain. Report services usually publish a wildcard that covers everyone. See external report authorisation.

Failure reports (ruf)

RFC 9991 keeps failure reports and the fo tag, and adds an Identity-Alignment field saying which mechanisms failed to align. It is also frank about their limits: a failure report can contain the headers or even the full content of someone’s message, so many large providers send few or none, and aggregate reports remain the recommended way to see what is happening. External ruf= addresses need the same authorisation record as rua=. If you do not have a specific reason to request failure reports, and a way to handle the personal data in them, leave ruf= out. See rua and ruf.

How to update your record, step by step

  1. Read what is published. Run your domain through the DMARC checker, and each subdomain that sends mail too. Note which record each one actually gets.
  2. Remove what DMARC no longer has. Delete pct=, rf=, ri= and any !size suffix on a report address.
  3. Decide about test mode. If you were ramping up with pct, choose between the full policy and t=y.
  4. Add np=reject unless you have a reason not to.
  5. Proofread. Check that it has v=DMARC1 first, exactly one record, and valid p, sp and np values.
  6. Check subdomain records. If a subdomain should keep its own policy for everything beneath it, give it psd=n.
  7. Check report authorisation for any rua= or ruf= address at another domain.
  8. Publish, then verify. Replace the TXT record at _dmarc, watch it propagate, and run the checker again.

A typical record written for RFC 7489:

v=DMARC1; p=quarantine; pct=50; rua=mailto:dmarc@example.com!10m; rf=afrf; ri=86400

Updated for RFC 9989, if your own mail passes and you want the policy in full:

v=DMARC1; p=quarantine; np=reject; rua=mailto:dmarc@example.com

Or, if you are still trialling quarantine:

v=DMARC1; p=quarantine; t=y; np=reject; rua=mailto:dmarc@example.com

The DMARC generator starts from the record you publish today, drops the tags DMARC no longer has, and offers test mode and np as options.

Frequently asked questions

Will my existing DMARC record stop working?

No, provided it is valid. Receivers ignore tags they do not know, so a leftover pct, rf or ri does no harm. The things that can go wrong are the ones in the mistakes table: a typo in p, sp or np, a lowercase v=dmarc1, or a pct you were relying on for a partial rollout.

Do all mailbox providers follow RFC 9989 already?

Receivers adopt a new standard at their own pace, so for a while some will follow RFC 9989 and some RFC 7489. A record without pct, rf and ri means the same thing to both, which is the best reason to clean it up now. The discovery_method field in your aggregate reports shows which receivers already use the tree walk.

Will older receivers honour t=y?

Not necessarily. To a receiver still on RFC 7489, t is an unknown tag, and it applies your policy in full. If you need a trial that every receiver honours during the transition, you can publish t=y and pct=0 together. Receivers following RFC 9989 read t=y, older ones read pct=0, and both apply one level below. Remove both when the trial ends. Our checker will still point out the pct; in this one case, that is expected.

Is p=none still enough for Gmail, Yahoo and Microsoft?

For their bulk-sender requirements, publishing a valid DMARC record is what they ask for, and p=none is a valid policy. It does nothing to stop anyone forging your domain, though, so treat it as step one of the rollout, not the destination. See p=none vs quarantine vs reject.

Why does my subdomain now show a different record?

Because of the tree walk. A subdomain without its own record now gets its organisational domain’s record, which is the shortest name above it with a record, not simply the nearest one. If an intermediate domain should keep its own policy, publish psd=n there.

Did SPF or DKIM change?

No. DMARCbis changes DMARC only. SPF is still RFC 7208 and DKIM is still RFC 6376. DMARC still passes when either one passes and aligns with the From domain.

Check your domain

Every tool on DNS Toolbox now reads DMARC the RFC 9989 way. The DMARC checker shows the record receivers apply, including one a subdomain inherits through the tree walk. It explains every tag, names the ones DMARC no longer has, and checks your report addresses are authorised. The DMARC generator writes a clean record with test mode and np. And if you want the reports read for you, a free account gives you a report address that turns the daily XML into a list of every service sending as your domain, with the next step for your policy.

The specifications

  • RFC 9989: Domain-based Message Authentication, Reporting, and Conformance (DMARC). May 2026, Standards Track.
  • RFC 9990: DMARC Aggregate Reporting. May 2026, Standards Track.
  • RFC 9991: DMARC Failure Reporting. May 2026, Standards Track.
  • RFC 7489 (2015, Informational) and RFC 9091 (2021, Experimental), both obsoleted by RFC 9989.

Checked against the published RFCs on 25 September 2026.

Check your own domain

Run a free scan of DNS, SPF, DKIM, DMARC, SSL and the blocklists, with the fix for everything it finds. Add your domain and we keep monitoring it, and email you the day something changes. Free for your first domain.

Create a free account