← all systems · see it in The System
Digital signatures and certificates
A digital signature is a mathematical scheme for verifying the authenticity of digital messages or documents: key generation, signing with a private key, and verifying with the public key, so that anyone can recognise the signature as authentic and only the signer can make it (src:wp:digital-signature; Diffie and Hellman 1976). It is 'an electronic analog of a written signature' that also detects whether data was modified after signing (FIPS 186-5). Walked here with the certificates that bind public keys to names, the certificate authorities that sign them and the chain of trust back to a trust anchor (RFC 5280), as the built system: schemes, keys, certificates, CAs, root programs, revocation, transparency logs, the laws that give signatures effect, its history and its people.
Properties published scienceour proposal
Sourced fields, and our own readings.
| aliases | Digital signature |
| born at | 1976 (notion published); 1977-78 (first scheme, RSA); 1989 (first widely marketed product, Lotus Notes 1.0) |
| born precision | year |
| boundary | Inside: signature schemes (key generation, signing, verifying) and their keys; the message digest; public key certificates and certification paths; certificate authorities, root programs, revocation, Certificate Transparency logs and webs of trust; the devices that hold signing keys (smart cards, HSMs); timestamps; the standards and laws that define and give effect to signatures. Outside: the documents, code and transactions that are signed, the networks that carry them, the computers that run the algorithms, the hash functions and hard problems it borrows, the people who sign and rely, and the attackers who forge. |
| nested level | Built from key pairs, hash functions, trap-door or lattice problems and certificates; embedded in public-key cryptography (one of its schemes), in public key infrastructure, and in authentication (one of its verifications); carried by computers, the Internet, people and law. |
| scale | technological |
Measures (25) published science
| characteristic clock: maximum TLS certificate lifetime (Baseline Requirements, from 15 March 2029) | 47 days |
| maximum TLS certificate lifetime, by date (Baseline Requirements) | 60 months (July 2012), 39 months (April 2015), 825 days (March 2018), 398 days (September 2020), 200 days (March 2026), 100 days (March 2027), 47 days (March 2029) lifetime |
| Let's Encrypt certificate lifetime | 90 days |
| issuance delay after the final click (RapidSSL, 2008) | 6 seconds |
| rogue-CA collision computation | about 18 hours on 200 PlayStation 3 consoles, 30 GB memory |
| CA validity periods seen (2013) | up to 40 years |
| breach detection to root removal (DigiNotar) | about 10 (19 June to about 29 August 2011) weeks |
| Ed25519 signature size | 64 octets |
| ML-DSA signature sizes (44 / 65 / 87) | 2420 / 3309 / 4627; public keys 1312 / 1952 / 2592 bytes |
| browser-trusted CA certificates (2013) | 1,832 controlled by 683 organisations certificates |
| trusted certificates controlled by three organisations (2013) | 75 percent |
| HTTPS sites behind one intermediate (2013) | 26 percent |
| hosts serving trusted certificates in a way that errors in some browser (2013) | 12.7 percent |
| hosts with port 443 open completing a TLS handshake (2013) | 67 percent |
| root certificates trusted by Firefox (2020) | 147 roots, 52 organisations (macOS 168/60; Windows 255/101) |
| certificates issued by Let's Encrypt to January 2019 | over 538 million, for 223 million domain names certificates |
| Let's Encrypt usage share (W3Techs, July 2024) | 60.6 (market share 64.1) percent of sites |
| participants who ignored the warning and used the bank site (FF2 / IE7) | 18 of 20 (90%) / 18 of 20 (90%) participants |
| properly signed malware samples still valid (2017) | 57.7 percent of 189 |
| certificates used to sign malware likely revoked over 15 years | about 20 percent |
| signed-malware amplification | 517 (180 developer machines to 93,016 machines) times |
| MD5-signed certificates in the rogue-CA team's sample (2008) | about 9,000 of 30,000 (97% from RapidSSL) certificates |
| quantum estimate to factor 2048-bit RSA | under a week with under a million noisy qubits qubits, time |
| jurisdictions adopting the UNCITRAL Model Law on Electronic Signatures | some 30 jurisdictions |
| UK adults with no consistent signature (2018) | a fifth (21% of 18-24) share of adults |
Open contentions (4) open contention
Where sources or readings disagree. Both stay on the map.
- C-1465 What was the first digital signature?
booked: The map keeps three firsts: the idea (1976), the first scheme (1977-78), and the first scheme proven secure.
the other reading: RSA 1977 as the first digital signature; GMR 1988 as the first that meets today's definition; Lamport, Merkle or Rabin one-time and tree schemes as first (1979, not read here) - C-1476 Can someone who digitally signed something deny it later?
booked: The map records that it's hard to deny only if the key stayed private and the signer saw what they signed, and keeps the stronger claims alongside.
the other reading: signatures give non-repudiation outright (RSA 1978: 'a signer cannot later deny'); non-repudiation is a legal, not a cryptographic, property - C-1478 What's the best way to decide whom to trust online?
booked: The map records the browser certificate system as it is used, and keeps the alternatives (web of trust, DNS-anchored keys) alongside.
the other reading: web of trust (PGP); DNS-anchored keys (DANE with DNSSEC), said to 'greatly reduce if not eliminate the role of trusted third parties'; transparency logs as the real safeguard rather than CAs - C-1479 Is an electronic signature as good as a handwritten one in law?
booked: The map records that it depends on the law and the kind of signature, and keeps the different legal readings.
the other reading: a digital signature is the legal equivalent of a handwritten one; a digital signature is evidence of key use only, not of a person's assent (WYSIWYS, key theft)
Cross-scale patterns seen here (23) our proposal
Shapes that recur when the same kind of problem is solved at different scales, and how each shows in Digital signatures and certificates. Each links to every other system that shows it.
- borrowed-then-held (structural)
Let's Encrypt borrowed trust (cross-signed by IdenTrust, October 2014), then held its own: ISRG's root accepted by all major root programs by August 2018, its own chain the default from July 2020. Earlier, RSA's 1978 directory key borrowed in-person presence before signatures carried the trust. - nested ladder (structural)
GMR's attack models (key-only, known-message, chosen-message, adaptive chosen-message) in increasing severity and breaks (total, universal, selective, existential) in decreasing severity: security at the top rung implies security at those below. The law repeats the ladder: electronic, advanced, qual - same role, different builds (structural)
One job, marking one's own so a checker can tell it from a stranger's, done by unrelated builds: the handwritten signature (a motor gesture), name-seals, thumbprints, the digital signature (a trap-door function), and in bacteria host-dependent DNA methylation, where the cell's own DNA carries the ma - same shape, opposite sign (structural)
One mechanism, the RSA trap-door permutation, run in the two directions: to encrypt, apply the recipient's public E and only D undoes it; to sign, Bob applies his private D_B (S = D_B(M)) and anyone undoes it with the public E_B (M = E_B(S)). Same permutation pair, the roles of public and private sw - seed (structural)
A 64-octet packet that, verified, lets any of n users trust any message from any signer with no prior pair contact; through ~150 roots, any client can trust any of hundreds of millions of names. The 1976 notion unpacked into signed software, HTTPS, legal e-signatures and 'an electronic coin as a cha - symmetry breaking (name the cut) (structural)
Before: shared-key authentication protects against third parties 'but do not protect against disputes between transmitter and receiver', since both hold the same key (a MAC 'only gives confidence to a specific recipient'). The cut: signing and verifying get different keys, so only the signer can pro - templating (structural)
One module reused unchanged: the X.509 certificate (1988) serves HTTPS, code signing, S/MIME, timestamping and OCSP signing (one key-purpose list in RFC 5280); the same signature standard serves all. Swapped algorithms keep the same certificate frame. - the whole sets the local reference (structural)
Each verifier measures trust against the anchors its root program installed; each reading is true locally, yet the readings do not close into one frame: 'There is no requirement that the same trust anchor be used to validate all certification paths', and in 2020 Firefox, macOS and Windows trusted 14 - trade one property for another (structural)
Kinds spread along trades: post-quantum security bought with size (Ed25519 signatures 64 octets, ML-DSA-44 2420 bytes); revocation timeliness against bandwidth and privacy (CRL, OCSP, pushed lists); fail-hard (security) against fail-soft (availability); PureEdDSA's collision resilience against HashE - who opens the door (structural)
The receiver opens the door ahead of time: the bank's central office 'can arrange beforehand to have a public key on file'; browsers ship root certificates; RSA's 1978 plan handed the directory's key over in person. A signature arriving finds a port already installed. Counter-case: a user who adds a - barrier-gates-complexity (partial)
The sealed private key is what everything above stands on; the without-it case is stated for the signature ('that party can produce perfect digital signatures of anything') and for the PKI ('the security of the entire system is lost'). Secure boot shows the build-up: hardware that boots only signed - causal ceiling (partial)
A limit set by the structure, not by means: 'neither public key cryptosystems nor one-way authentication systems can be unconditionally secure because the public information always determines the secret information'. Security is computational by construction; the structural form (information-theoret - conservation law plus closure (partial)
Verification is a law with an open slot: it proves the holder of some private key signed, not who that holder is. A second rule closes the slot: 'This assurance can only be provided if the owner's identity and public key are bound together, such as in a certificate'. Without the closure, 'forging a - holding / entrainment (partial)
The whole (root programs through the Baseline Requirements) sets when every certificate dies: maximum lifetimes stepped from 60 months to 398 days and, by ballot, to 47 days; Let's Encrypt sets 90 days for its own. - lightning (partial)
A stored difference released at once: one MD5 collision (18 hours on 200 PS3s) yields a CA certificate 'trusted by all common web browsers'; one breached CA (DigiNotar) issues hundreds of certificates and is bankrupt within three months. After the discharge the system is not the same (MD5 dropped, t - namespace completeness (the map predicts its gaps) (partial)
GMR's grid of attack models against kinds of break predicted its hardest empty slot, security against existential forgery under adaptive chosen-message attack, which folklore held impossible ('considered to be contradictory'); GMR filled it in 1988. - outside-to-inside (partial)
The partition between trusted and untrusted keys first lay outside the system (showing up in person; a notary who must 'personally know' the signer), then moved inside it, into the root store and certificate chain every client carries. - red-queen (partial)
Designers and forgers both improve (MD5 to SHA-2, DSA retired, ML-DSA; collisions, stolen keys, quantum estimates) and the system keeps moving to stay trusted. The stalemate is read in the bacterial analogue's source, not stated for signatures: restriction-modification diversity read as the Red Quee - same toolkit, independent builds (partial)
From the same parts (signatures over public keys), key-to-name binding was built as a top-down hierarchy (X.509, 1988) and as a web of trust (PGP, 1992), 'in a format that performs a similar function'. Shared ancestry in public-key cryptography is likely, so independence is partial. - The carry (partial)
The wait face: a logged certificate waits up to the log's Maximum Merge Delay before it is in the tree; revocation waits for clients to fetch CRLs or ask OCSP (latency and privacy costs). Short lifetimes are a way round the wait: a compromised certificate lapses rather than waiting on revocation to - The Payment to Entropy (partial)
Two forms. Loss by design: signing the condensed digest drops distinctions on purpose, and the dropped part returns as collisions (two certificates, one MD5 digest, one valid signature). A deployment toll per hand-off: of hosts with port 443 open only 67% completed a TLS handshake, and 12.7% of host - Umwelt: each system carves its own slice of one shared medium (partial)
Each verifier sees the shared certificate world only through its own root store and checks (Firefox 147 roots, Windows 255; some clients skip revocation); people see the screen but not the bits they sign ('the user does not see what they sign'). - partner-becomes-part (surface)
Certificate Transparency logs stay separate operators, but their promises (SCTs) are embedded in the final certificate as an extension: a partner's token carried inside, with separate identities kept (the dark twin's form, booked as surface).
Valence: what it trades, and with whom (16) published science
Each exchange with a neighbour, read from both sides: who gains, who pays, and what carries it.
- Cryptographic hash function via sys:written-record · Digital signatures and certificates mixed
- Human (Homo sapiens) via sys:sensory-environment · Digital signatures and certificates mixed
- Writing (the practice of marking language on a surface, and the script plus rules it works with) via sys:written-record · Digital signatures and certificates gain
- Law (rules and law) via sys:law · Digital signatures and certificates gain
- Internet via sys:electromagnetic-field · Digital signatures and certificates mixed
- Money via sys:written-record · Digital signatures and certificates gain
- Bitcoin (cryptocurrency) via sys:written-record · Digital signatures and certificates gain
- Operating system via sys:written-record · Digital signatures and certificates gain
- Quantum computing via sys:written-record · Digital signatures and certificates loss
- Public-key cryptography via sys:written-record · Digital signatures and certificates gain
- Cryptanalysis (the attack side of the practice) via sys:written-record · Digital signatures and certificates mixed
- Authentication via sys:written-record · Digital signatures and certificates gain
- Threat actor via sys:written-record · Digital signatures and certificates loss
- Protocols: TLS (and SSH, PGP, S/MIME) via sys:electromagnetic-field · Digital signatures and certificates gain
- Supply chain security via sys:digital-signature · Digital signatures and certificates gain
- Digital identity via sys:human-society · Digital signatures and certificates gain
Partners and parts published science
Built from / builds (13)
- Public-key cryptography
- Cryptographic hash function
- Key pair (public key and private key)
- Certificate transparency logs
- Certificate authorities (trust anchors)
- Signature algorithm
- Public key certificate
- Root program
- Certificate revocation
- Trusted timestamping
- Hardware security module
- Smart card
- Authentication
Sits inside / holds (3)
expressed by (24)
- Generation/Synthesis
- Recombination/Composition
- Templating / Standardization
- Interoperability/Transduction (Bridging)
- Encapsulation & Modularity (Abstraction)
- Amplification/Scale-Up (Manufacture & Distribution)
- Distributed Autonomy/Decentralized Execution (Subsidiarity)
- Internal Coordination/Regulation
- Sensing/Scanning
- Uncertainty Reduction/Learning
- Trail-Marking / Externalized Memory (Stigmergy)
- Surprise / Change-Point Detection
- Reactive Response
- Proactive Defense
- Physical/Virtual Barrier
- Self-Monitoring
- Validation of Identity/Intention
- Validation of Effectiveness
- Enclave/Segregation
- Resilience & Repair
- Triage & Prioritized Allocation
- Quality Control & Error Correction
- Positional Identity (Body Map)
- Scheduled Build-and-Dismantle (On-Demand Structure)
floor filled by (22)
- Cryptographic hash function
- Public-key cryptography
- Human (Homo sapiens)
- Written record (durable external marks: texts, archives, canon)
- Law (rules and law)
- Internet
- Computer
- Operating system
- Authentication
- Key pair (public key and private key)
- Certificate transparency logs
- Web of trust
- Random number generator
- Certificate authorities (trust anchors)
- Signature algorithm
- Public key certificate
- Root program
- Certificate revocation
- Trusted timestamping
- Hardware security module
- Smart card
- Digital identity
represents (1)
Kinds (16)
- Public-key cryptography
- Web of trust
- Hash-based signatures (Lamport, Merkle, SPHINCS+, SLH-DSA)
- RSA signature
- Digital Signature Algorithm
- ECDSA
- EdDSA (Ed25519, Ed448)
- ML-DSA
- Falcon
- Schnorr and ElGamal signatures
- BLS (pairing-based) signatures
- GMR signature scheme
- Quantum digital signature
- Code signing
- DANE
- Electronic signature
wired (1)
Notes published science
- In 2008 a team turned about 200 PlayStation 3 consoles into a collision machine (about 18 hours of computing) and got a certificate authority certificate that every common browser trusted. The trick worked because one CA issued certificates 'exactly 6 seconds after the user clicks', so they could predict what it would sign. (src:sotirov-2008-rogue-ca)
- Digitally signed malware was already in the wild in 2003. Copying a legitimate file's signature onto known malware made some anti-virus products stop flagging it, even though the signature didn't match the file, and only about 20% of the certificates used to sign malware were likely revoked over 15 years. (src:kim-2017-certified-malware)
- Bacteria sign their own DNA. Restriction-modification systems methylate the cell's DNA at set sites and cut DNA that lacks the mark: modifications 'used by cells to distinguish self from non-self'. Viruses that escape the cut evolve fewer of those sites, and the review reads the diversity of these defences as a Red Queen race. (src:anton-2025-rm)
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
digital signatures and certificates: signed bits up to evidence a stranger can rely on (reader's view)
- L1: Signed data in the world: documents, e-mail, software updates, certificates, transactions; any bitstring, stored or sent
- L2: The message digest: the data condensed by a hash function (EdDSA hashes inside the algorithm)
- L3: The key pair: a private signing key held by the signatory (smart card, HSM) and a public verification key anyone may hold
- L4: Signing and verifying: RSA, ECDSA, EdDSA, ML-DSA; verification accepts or rejects; the security notion is unforgeability under adaptive chosen-message attack
- L5: Binding a key to a name: the certificate (X.509) and the certification path from the leaf through intermediates to a trust anchor, checked for validity, constraints and revocation
- L6: Governing trust: certificate authorities as trusted third parties, root programs and the CA/Browser Forum's Baseline Requirements, revocation, Certificate Transparency logs, webs of trust, and the law that gives signatures effect
- L7: What it gives people and what they do with it: authenticity, integrity and non-repudiation, evidence a third party (a court) can rely on; acting on it by installing the update, opening the connection, accepting the contract or the coin
the boardroom (shared floors)
- L4: Signature generation and verification: the algorithms (RSA, ECDSA, EdDSA, ML-DSA) and their security notion
- L5: Certification path validation: binding the key to a name, from leaf to trust anchor, with validity, constraints and revocation
- L6: Trust governance: CAs, root programs and the Baseline Requirements, Certificate Transparency, webs of trust, law
- L7: Assurance and action: authenticity, integrity, non-repudiation; evidence a third party can rely on, acted on by people and machines
way in: Attacks and research results: collisions, forgeries, key-recovery, quantum estimates
- L1: Published results: weaknesses in hashes and keys, quantum resource estimates
- L2: Proofs of concept against the deployed system: a rogue CA certificate from an MD5 collision
- L3: Measured against what is deployed: MD5 still signed until 2013, 1024-bit RSA in half of chains; current standards not quantum-resistant
- L4: Reaches the verify step: the attack models (key-only to adaptive chosen-message) and breaks (existential forgery to total break)
- L5: The path's weakest link: any trusted CA can sign for any name, so one break opens all
way in: Certificate Transparency: logs, signed timestamps and proofs
- L1: Publicly auditable, append-only logs of issued certificates
- L2: A Signed Certificate Timestamp: the log's promise to merge within the Maximum Merge Delay
- L3: Merkle audit and consistency proofs show the log only grows
- L4: Auditors check that each certificate with an SCT is in the log
- L5: Monitors and domain owners notice suspect certificates; a log's misbehaviour is itself evidence
way in: Certificates presented: the leaf and its intermediates
- L1: The server or signer presents a leaf certificate with the chain linking it to a trusted root
- L2: Parsed: subject, issuer, validity period, public key, extensions, the CA's signature
- L3: A path assembled from leaf to anchor; often incomplete or misordered, and the global directory to fetch missing links never materialised
- L4: Each certificate's signature checked with its issuer's public key
way in: Identity evidence brought to a CA by applicants
- L1: An applicant asks for a certificate for a name
- L2: Domain validation probes (email to administrative addresses, ACME challenges); EV adds identity documents
- L3: The probes cross unprotected networks and can be watched or hijacked
- L4: The CA decides to sign the binding, fully robotically at Let's Encrypt
- L5: Bound by the Baseline Requirements, a condition of root-store inclusion
way in: Law and policy: statutes, model laws and regulations
- L1: Statutes enacted, Utah first in 1995; the UNCITRAL model laws (1996, 2001); ESIGN (2000); eIDAS (2014)
- L2: What a signature binds: anything digitally signed legally binds the signer to its terms, where the statute says so
- L3: Tiers of signature: electronic, advanced, qualified (certificate, secure creation device, qualified trust service provider)
- L4: A qualified signature is hard to challenge in court
- L5: Accreditation ties CAs to local law, so the CA business is national
way in: Randomness: key-generation seeds and per-message secret numbers
- L1: An entropy source feeding a random bit generator
- L2: Random bits drawn into a key pair or a per-message secret number k
- L3: Its quality matters per signature: weak randomness in k reveals the private key; EdDSA is deterministic; ML-DSA hedges fresh and precomputed randomness
way in: Revocation status: CRLs, OCSP answers, pushed lists
- L1: The issuing CA's cryptographically authenticated statement that a certificate is revoked
- L2: Fetched as a CRL or asked of an OCSP responder
- L3: Timeliness traded against bandwidth, latency and privacy
- L4: Applied, or skipped: browsers limit checks and fail soft
way in: Signed data arriving: messages, documents, updates, transactions with their signatures
- L1: The data and its signature as received or retrieved
- L2: The verifier recomputes the digest with the same hash function
- L3: The claimed signatory's public key is taken up
- L4: Verification accepts or rejects the signature
way in: The person: the signer's intent and the relying party's eyes
- L1: Indicators and warnings on the screen; a card and a PIN pad in the hand
- L2: Seen, but the bits are not: the user does not see what they sign
- L3: Risk perception decides whether to heed
- L4: Heed or click through: most ignored the FF2 and IE7 warnings
- L5: The person's own trust settings: adding or removing CA certificates
- L6: What signing means to the signer: identity, intent and informed consent
way in: Time: the verifier's clock, validity periods and trusted timestamps
- L1: The verifier's current time and, for code, a timestamp from a timestamping authority
- L2: A signed timestamp binds the signature to a moment
- L3: Compared with the certificate's notBefore and notAfter
- L4: The certificate is accepted only within its window; timestamped code outlives revocation
way in: Trust anchors: the root store installed ahead of time
- L1: Root certificates shipped with browsers and operating systems by their vendors (in RSA's 1978 plan, the directory's key handed over in person)
- L2: The root store: self-signed roots, plus intermediates cross-signed into trust
- L3: Policy selects the anchor for a path
- L4: The anchor's key verifies the top of the chain; different verifiers may use different anchors
way out: A CA distrusted: its roots removed from root stores
- L1: The roots leave the shipped stores; the CA's business can end
- L6: Root programs judge a CA compromised or misbehaving
way out: A certificate issued
- L1: The certificate delivered to the subject and served to relying parties
- L5: Signed with an intermediate key held in an online HSM
- L6: The CA certifies the binding under its policy
way out: A certificate logged for transparency
- L1: The log grows by one entry, for anyone to read
- L6: The CA submits the (pre)certificate to one or more logs
way out: A certificate revoked
- L1: A signed revocation statement in a CRL or OCSP answer
- L5: The CA publishes revocation status
- L6: Compromise, error or a changed binding: the CA must revoke
way out: A signature made and attached
- L1: The data goes out with its signature
- L3: The private key stays in the card or HSM
- L4: The signing algorithm runs on the digest with the private key
- L5: The key is unlocked: a smart card activated with a PIN
- L7: A person or program intends to sign: approval, consent, a release
way out: A warning shown to a person
- L1: A warning page, reaching the person's sight
- L7: Validation failed or is in doubt
way out: Acted on: the update installed, the connection opened, the payment or contract accepted
- L1: The code runs, the session proceeds, the coin moves
- L5: Each layer runs the next only if its signature checks
- L7: Accepted as authentic
way out: Algorithms retired and replaced
- L1: Published as a standard
- L4: New algorithms specified for signing
- L6: Standards bodies retire what attacks or quantum risk have weakened