You receive an invoice email and your mail service reports that its signature is valid. There is no handwritten name and no signature image. What has it checked?
The service has checked information that the sending system added so another system can verify it. That information is not a picture of a signature. It is the result of a calculation tied to particular message content and to secret signing information held by the sending service. We will follow those steps before considering what a successful result means. 4
A seal that the receiving service can check
An email address such as [email protected] contains a domain name after @: example.com. A business can arrange for its mail provider to send messages and apply a checkable mark associated with a domain. The provider’s mail server is the computer running the software that performs this work. You write the invoice notice; its sending software adds the mark as part of preparing the message for delivery. The mechanism is called DKIM, short for DomainKeys Identified Mail. It associates a signing domain with selected parts of the message, rather than certifying the individual who typed it. 4
Think of DKIM as a tamper-evident seal applied by a named organization. In the actual email check, that means checking the signing domain and the covered content, not independently identifying a legal company. It does not tell you whether the person who prepared the contents made an honest or correct decision.
The sending service holds a private key: secret information used to create signatures. A corresponding public key is published in DNS, the Internet’s directory for domain information. The receiving service fetches that public key and uses it to verify the signature. The public key is meant to be available; knowing it does not provide the private key’s ability to sign. 4
Here, a key is data used by a calculation, not a password the customer must type. The two keys have different jobs. The private key gives the sending system the ability to create the mark. The public key gives the receiving system a way to test it without obtaining that signing ability. This separation lets many recipients verify mail without the sender sharing its signing secret with them. 4
The sending software first calculates a compact value from the message body: the content after its introductory information lines. That value is called a hash. It is derived from the content, not a readable copy of it.
Next, the software uses its private key to sign the specified message information, including that body value. It attaches a special information line containing the signature and the details needed for checking.
The receiving software repeats the relevant calculations on the received content and uses the public key to verify the signature. If covered content changes beyond what the signing rules tolerate, the checks no longer fit. The receiver is checking the received material against the signed material, not reading an image of a seal. 4
For our invoice notice, imagine a covered line in the message changing from the original payment instruction to a different one after signing. A successful check should not survive that covered change under the signing rules. But if the false instruction was already there when the service signed, the receiver’s calculations can agree perfectly. The check compares the received material with what was signed; it does not compare the instruction with the supplier’s accounting records. 4
The signature covers selected message information, called headers, and the message body according to the chosen signing rules. Those rules can tolerate harmless changes in spacing, and a signature can specify a body-length limit. So “DKIM passed” should not be expanded into “every displayed detail and every byte is guaranteed unchanged”. It means the covered material passed the defined check. 4
DKIM does not hide the message. Signing and encryption do different jobs: a signature enables verification, while encryption protects readability against those without the necessary access. Publishing a DKIM key therefore does not make an invoice confidential. 4
How the receiver finds the right key
A business can send ordinary staff mail through one provider and invoice notifications through another. It may also replace old signing keys with new ones. The receiver therefore needs more than the business’s domain name to know which public key to fetch. 4
DNS, where the public key is published, is a directory: software asks for information at a particular domain name. It does not search every possible key label until something works. The signature must tell the receiver where to look. Otherwise, a company using separate keys for staff messages and invoices would leave the receiver unable to choose the relevant one just from the company name. 4
The signature carries two useful pieces of information. One is the signing domain, marked d=. The other is a key label called a selector, marked s=. Together they identify where the public key should be looked up. For example, the domain example.com and selector billing point to billing._domainkey.example.com. You do not need to memorize that name; the important part is that the key label is part of the address. 4
This explains a misleading result in some domain checks. If a tool knows only example.com and guesses several common selector names, finding nothing does not establish that the business has no DKIM. It may simply be looking under the wrong labels. With an unknown selector, the accurate conclusion is “DKIM was not verified from the available information”, not “DKIM is absent”. 4
The DKIM signature in an original received email supplies the actual domain and selector. A public key found at that location is useful evidence of configuration, but checking the message’s signature is a further step. A key can be published even if a particular application is not using it. 4
What to make of a pass or a failure
For an invoice notification, the most useful result comes from the service that actually received and checked it. Mail systems can record their conclusions in a field called Authentication-Results. That field is meaningful when it comes from a receiving system you trust; anybody’s pasted text saying “pass” is not equivalent evidence. An administrator can distinguish the trusted receiving service’s result from other text in a message. 11
A failure deserves explanation. An intermediate system may have changed signed material in a way the signature does not tolerate, or the required key information may not be usable. It is not automatically proof that an attacker changed the invoice. Looking at the original received message is more useful than inspecting a newly composed forward that does not preserve the original structure. 4
A pass also needs the signing domain beside it. A service can validly sign with its own domain rather than the business’s domain. DMARC, the mechanism discussed next, checks whether a successful signature’s domain is sufficiently connected to the domain visible in From. Any valid signature is not necessarily the right signature for that purpose. 6
Why a signed message can still be dishonest
Consider a supplier account taken over by a criminal. If the criminal sends through the supplier’s normal service, that service may sign the message just as it signs genuine mail. The signature can be valid even though the request to change a bank account is fraudulent. This follows from what DKIM checks and from the compromised-account fraud described by the FBI. 4 12
The seal analogy now has its limit in plain sight. The message can pass through the genuine sealing process while the person using that process should not be trusted. DKIM remains valuable for checking domain-associated message material; it simply cannot make the supplier’s commercial decisions.
For a small business, the sensible division is straightforward: let the mail service check the signature, let its administrator explain any unexpected result, and verify consequential payment changes through an established independent contact. A valid signature is a reason to understand the technical origin better, not a reason to skip that last conversation.