Skip to content
IAM Insights

Explainers · Glossary

ACME

Automatic Certificate Management Environment · ACME DNS-01 · ACME HTTP-01

By Boston Identity · Explained · 7 min read

The short answer

ACME is the IETF protocol that lets software get a certificate from a certificate authority with no person involved. The authority sets a challenge that only someone controlling the domain can complete, then issues the certificate.

What it is
A protocol
Defined by
IETF, RFC 8555 (standards track, 2019)
Who it's for
Teams that run certificates, PKI or machine identity
In our briefings
9By month of the source: 2025-12: 0, 2026-01: 1, 2026-02: 0, 2026-03: 0, 2026-04: 2, 2026-05: 0, 2026-06: 0, 2026-07: 0, 2026-08: 0, 2026-09: 0, 2026-10: 4, 2026-11: 11by month, last 12
01

What is ACME?

ACME, the Automatic Certificate Management Environment, is an Internet Engineering Task Force (IETF) protocol, published as RFC 8555 in March 2019 on the standards track.S1 It automates how a server gets a certificate from a certificate authority (CA). A certificate is the CA's statement that a key belongs to a name.§10

RFC 8555 describes what came before as ad hoc. A person generated a certificate signing request (CSR), pasted it into a CA's web page, proved ownership of the domain by following written instructions, then downloaded the certificate and installed it. The authors' informal tests found that webmasters often needed one to three hours to get and install a certificate for a domain.§1

ACME turns those steps into JSON messages (JavaScript Object Notation, a plain-text data format) carried over HTTPS.§4 Two roles exchange them. The client runs on the server that needs the certificate, or on a separate server authorized to answer the CA's challenge. The ACME server runs at the CA.§3 The RFC's focus is domain names in the Web PKI (public key infrastructure), and it says ACME supports extensions for other identifiers.§1

Not the same as proof of who the requester is. ACME's domain validation checks that the requester has effective control of the domain, and RFC 8555 says the CA is not required to verify the requester's real-world identity.§1

02

How does ACME work?

Two parties speak ACME: the client, which acts for whoever runs the domain, and the ACME server at the CA. A third place matters as well. The CA makes its own outbound requests to the domain's web server or its DNS (Domain Name System) records, on what the RFC calls the validation channel.§10.1

Figure steps
  1. The client opens an account
  2. The client orders a certificate
  3. The server sets challenges
  4. The client publishes its proof
  5. The server checks for itself
  6. The client finalizes and collects
Fig. 1

The certificate authority never takes the client's word for it: it reads the proof from the domain itself, over a separate connection, before it issues anything.

Drawn from RFC 8555, sections 4, 7 and 8

Swipe for the full figure →

  1. The client opens an account

    The client generates a key pair and requests a new account. It signs the request with the private key to prove it controls that key, and it can add contact details and agree to the CA's terms of service.§4 From then on the client signs all its messages with this account key, and the server checks them with the public key.§3

  2. The client orders a certificate

    The client posts an order to the server's newOrder address, listing the identifiers it wants in the certificate, here domain names. If the server is willing to issue, it replies with an order object. That object lists the authorizations the client must complete, and it carries an expiry time for completing them.§7.4

  3. The server sets challenges

    The client downloads each authorization.§7.5 Each one holds a set of challenges, the tests the server will accept for that identifier.§8 RFC 8555 defines two for domain names. The http-01 challenge asks for a file on the domain's web server.§8.3 The dns-01 challenge asks for a record in the domain's DNS.§8.4

  4. The client publishes its proof

    The client builds a key authorization, which is the challenge's random token joined to a fingerprint of the account key.§8.1 For http-01 it serves that string on the domain at a path under /.well-known/acme-challenge/.§8.3 For dns-01 it puts a digest of the string in a TXT record, a DNS text record, under the _acme-challenge label of the domain.§8.4 It then posts an empty JSON body to the challenge address to say it is ready.§7.5.1

  5. The server checks for itself

    The server builds the same key authorization from the token and the client's account key. For http-01 it fetches the file with an HTTP GET request, which must go to port 80, and compares the body with its stored copy.§8.3 For dns-01 it queries the TXT records for the validation name and looks for one that matches the digest.§8.4

  6. The client finalizes and collects

    Once the client believes the requirements are met, it posts a CSR to the order's finalize address. The CSR must name exactly the identifiers in the original order. When the order's status becomes valid, the server has issued the certificate.§7.4 The client downloads it from the certificate address, and to renew it later the client starts a new order.§7.4.2

03

Why does ACME matter now?

We read this year's drafts as stretching ACME from domain names to devices. An IETF draft adds two identifier types: one for a manufacturer-assigned identity, typically a serial number, and one for the secure cryptoprocessor that generated the certificate key. It also adds a challenge, device-attest-01, that proves control of them. The client answers with a WebAuthn attestation object made over the ACME key authorization or, for some formats, over the token alone.1 The draft is not yet an RFC. The IETF datatracker shows it approved for publication as a Proposed Standard and with the RFC Editor.1 Its authors say the extension is primarily useful in enterprise PKI.1

The same draft updates RFC 8555. For the two device identifier types, the CSR should repeat the identifiers in the order, where RFC 8555 says must, and clients may omit them.2 Revision -10, posted in August 2026, advises implementers to treat leaving device identifiers out of certificates as the default posture, unless there is a specific operational requirement for them to appear.2 One of the draft's reasons is that an identifier in a certificate logged to certificate transparency becomes, in its words, "permanently and publicly discoverable".2

ACME's challenge is also named in agent identity work. Cisco's draft for registry-signed agent cards says the issuer must not list endpoints at an origin the registrant has not been shown to control. A challenge such as ACME's is one of the three ways it accepts to show that control.3 By contrast, a draft certificate profile with an AgentIdentity extension, at its fifth revision, does not cite ACME.4

04

Where does it fit?

Compared withHow ACME differs
X.509 PKI Certificate and CRL Profile (RFC 5280)RFC 5280 is the document RFC 8555 cites for certificates themselves and for the reason codes used in revocation.§1,§7.6 ACME is the protocol around the certificate: the request, the proof of control and the download.§4
SPIFFEACME is a way to get a certificate; SPIFFE (Secure Production Identity Framework For Everyone) supplies an identifier that can go in one. The ACME device-attestation draft says implementers should prefer logical identifiers such as SPIFFE URIs over device identifiers in workload mutual TLS (Transport Layer Security), and treat attestation as a bootstrap only.2
05

What goes wrong?

  • ACME's challenges test control of the domain's DNS and web content, so whoever has that control can pass. RFC 8555 calls hosting providers a particular risk: if the domain owner has outsourced DNS or web services, nothing can be done against tampering by the provider.§10.2
  • Misconfiguration can hand that control to the wrong party. On a web server that lets non-administrative users write to .well-known, any user can claim the server's hostname by answering an HTTP challenge. If the ACME server follows redirects, a catch-all redirect rule can let the redirect's target get a certificate.§10.2
  • DNS is a common point of vulnerability for both challenges. An entity that can provision false DNS records can attack the DNS challenge directly, or direct the server's HTTP validation query to a server of the attacker's choosing.§10.2 The RFC recommends that the server query and connect from multiple points in the network.§10.2 It also recommends that CAs make DNS queries through resolvers that validate DNSSEC (DNS Security Extensions).§11.2
  • A stolen account key costs more than a stolen certificate key. The certificate key lets an attacker impersonate the names in that certificate for its lifetime. The account key lets the attacker issue certificates using existing authorizations, revoke existing certificates and lock out the legitimate account holder.§11.1
  • RFC 8555 says ACME's controls on issuance are focused on validating that the applicant controls the identifier. It lists other checks a CA might need to perform before issuing, such as whether the client has agreed to a subscriber agreement.§10.5
06

What should a security leader ask?

  1. 01

    Which of our certificates are issued through ACME, and which still depend on a person pasting a request into a CA's web page?

  2. 02

    Who can change our DNS records or write to the .well-known path on our web servers, hosting providers included?

  3. 03

    Where is each ACME account key stored, who can use it, and is each key used for one account only?

  4. 04

    Does each CA we use validate from multiple network points and through a DNSSEC-validating resolver?

  5. 05

    If we adopt ACME for devices, will device identifiers appear in the certificates, and will the CA require external account binding?

07

Sources

›Evidence: 4 claims from 3 briefings
  1. 1BI-FIND-2026-1050 · ACME's device-attestation draft now advises leaving device IDs out of certificates and drops a key-export ruleACME validates control of identifiers such as domain names, and this draft extends it to devices. It adds two identifier types, permanent-identifier for a manufacturer-assigned identity, typically a serial number, and hardware-module for the secure cryptoprocessor that generated the certificate key, plus a challenge, device-attest-01, that proves control of them. The client answers with a WebAuthn attestation object made over the ACME key authorization or, for some formats, over the token alone. The server runs WebAuthn's verification steps, checks the signed value against what it stored, and checks that the attested device identifier matches the order. The authors say the extension is primarily useful in enterprise PKI, recommend that the server require external account binding when an account is created, and name Android Key Attestation, Chrome OS Verified Access, the Trusted Platform Module and Apple's Managed Device Attestation as examples of the attestation devices already offer. The datatracker shows it approved for publication as a Proposed Standard and with the RFC Editor.
  2. 2BI-FIND-2026-1052 · ACME's device-attestation draft now advises leaving device IDs out of certificates and drops a key-export ruleThe ACME device-attestation draft now advises leaving device identifiers out of certificates. At -01 that was only an option: a client MAY include the permanent identifier in its certificate request, and a CA MAY leave the hardware-module identity out. Revision -10, posted in August 2026, adds a privacy section saying implementers "should treat this privacy-preserving mode as the default posture unless there is a specific operational requirement for the identifier to appear in the certificate." That "should" is lowercase, which makes it advice. The protocol change is an update to RFC 8555: for these two identifier types the certificate request SHOULD, rather than MUST, repeat the identifiers in the order, and clients MAY omit them. The draft's reasons are that an identifier in a certificate logged to certificate transparency becomes "permanently and publicly discoverable", and that in workload mutual TLS it ties identity to the physical device, so implementers should prefer logical identifiers such as SPIFFE URIs and treat attestation as a bootstrap only. The authors allow that identifiers in certificates may be acceptable in closed enterprise environments.
  3. 3BI-FIND-2026-1956 · Cisco's SD-Card rewrite pins MCP tools to what a registry signed; its security sections still read TODORevision -00 specifies how a registry issues a card, which our briefing on the -02 text found left out of scope. A registrant sends a request naming the binding, a metadata URL, its public key and the discovery contexts it wants, and proves it holds the key: with DPoP when it authenticates with an OAuth access token, or by another mechanism such as a WIMSE proof token. The issuer MUST NOT list endpoints at an origin the registrant has not been shown to control, shown by a client identifier URL at that origin, a challenge such as ACME's, or the origin's metadata document carrying the key. In the A2A binding the issuer's signature over the card takes the place of the Agent Card's own signatures member. The text still leaves out how a registrant finds a registry's registration endpoint and what the registry's vetting policy is, and the verification steps leave the choice of which registries to trust to "a pre-established trust relationship".
  4. 4BI-FIND-2026-0750 · Sharif's profile writes an AI agent's rights to pay and delete into an identity certificate valid up to a yearAt its fifth revision the draft is still internally inconsistent: it gives the AgentIdentity extension two different identifiers, reuses one identifier for both a key purpose and a branch of extensions, and adds a jurisdiction section missing from its contents that points to a section that does not exist. Every identifier sits under the author's company's enterprise number, assigned in July 2026. No message on the certificate working group's list or the other IETF lists we checked names the draft, and it does not cite WIMSE, ACME or OAuth's certificate-bound tokens.

›In our briefings

What our briefings say about ACME itself, with the most direct claims first.

  1. ACME validates control of identifiers such as domain names, and this draft extends it to devices. It adds two identifier types, permanent-identifier for a manufacturer-assigned identity, typically a serial number, and hardware-module for the secure cryptoprocessor that generated the certificate key, plus a challenge, device-attest-01, that proves control of them. The client answers with a WebAuthn attestation object made over the ACME key authorization or, for some formats, over the token alone. The server runs WebAuthn's verification steps, checks the signed value against what it stored, and checks that the attested device identifier matches the order. The authors say the extension is primarily useful in enterprise PKI, recommend that the server require external account binding when an account is created, and name Android Key Attestation, Chrome OS Verified Access, the Trusted Platform Module and Apple's Managed Device Attestation as examples of the attestation devices already offer. The datatracker shows it approved for publication as a Proposed Standard and with the RFC Editor.

    ACME's device-attestation draft now advises leaving device IDs out of certificates and drops a key-export ruleBI-FIND-2026-1050

›Cited in 9 briefings

Every briefing that discusses this term, newest first.

  1. Agent registry draft lets another identity block an agent's key changes, not switch it off

    Draft · · IETF Datatracker · Christopher Drake, 1id.com · Oct 27, 2026

  2. Cisco's SD-Card rewrite pins MCP tools to what a registry signed; its security sections still read TODO

    Draft · · IETF Datatracker · Suhas Nandakumar et al. · Nov 1, 2026

  3. An unadopted AGTP draft would let gateways refuse an agent's out-of-scope requests from its TLS certificate

    Draft · · IETF Datatracker · Chris Hood, Nomotic, Inc. · Oct 30, 2026

  4. Apps staff build with AI app builders leave a trail in certificate logs and DNS, a Cerebras security lead says

    Video · · OWASP Foundation · Uploaded October 21, 2026

  5. Sharif's profile writes an AI agent's rights to pay and delete into an identity certificate valid up to a year

    Draft · · IETF Datatracker · Raza Sharif, CyberSecAI · Oct 2, 2026

  6. The Agent Name Service draft was rebuilt on DNS, with a GoDaddy author, and no standards body has adopted it

    Draft · · IETF Datatracker · Scott Courtney et al. · Apr 13, 2026

  7. An agent certificate draft ties access to a trust score set by the issuing CA, usually the agent operator's own

    Draft · · IETF Datatracker · Raza Sharif, CyberSecAI · Apr 10, 2026

  8. ACME's device-attestation draft now advises leaving device IDs out of certificates and drops a key-export rule

    Draft · · IETF Datatracker · Brandon Weeks et al. · Jan 29, 2026

  9. The 2024 'IDs for AI' paper puts the ID on a chat session, not a model, and leaves trust in its issuer to others

    Paper · · arXiv · Alan Chan et al. · Jun 17, 2024

In our columns

  1. Hardware attestation proves a genuine chip, not that the server is yours

    Column · Oct 17, 2026

Was this explainer useful?

Share this explainer

Every finding, forecast and open question on this site has a permanent ID. If we reword one, its ID stays the same, so a citation keeps pointing to the same claim.

Follow us on LinkedInRSS feed

Working on an identity problem that involves ACME? Talk to Boston Identity