Glossary

Thirty terms that any security team working on financial infrastructure runs into, with a short explanation and a link to the article that goes deeper on each one.

The glossary is arranged in four groups by domain — networks and identity, keys and cryptography, trading infrastructure, and regulation and compliance — and within each group the order is thematic rather than alphabetical, so that related terms sit next to each other. The four categories follow the four layers I meet in every architecture review: who talks to whom, where the keys are kept, what happens on the trading side, and what the regulator expects to know about all of it. Every entry has a precise definition, one sentence explaining why it matters in financial infrastructure, and a link to further reading in the article that goes deeper; a browser search (Ctrl+F) works well because every entry has a fixed anchor.

Networks & identity

mTLS Mutual TLS
Mutual authentication at the TLS layer: the client presents a certificate of its own and the server verifies it, not only the other way round. Inside internal infrastructure this turns every request into the carrier of an explicit service identity, instead of an anonymous request arriving from a familiar IP address. Why it matters: without mTLS any process inside the network can impersonate another service; with mTLS in strict mode an attacker who takes one pod gets one door rather than a whole corridor. The protection breaks the moment someone adds a "temporary" bypass rule to the firewall. Read more →
TLS 1.3 Transport Layer Security 1.3
The current version of the TLS protocol, defined in RFC 8446: a single round-trip handshake, removal of ageing algorithms such as RSA key exchange and CBC, and Forward Secrecy as a requirement rather than an option. Why it matters: a connection left on TLS 1.2 with a weak cipher allows downgrade and retrospective listening; on TLS 1.3 the shorter handshake also removes latency from every transaction, and that is measured in milliseconds. Read more →
Zero Trust Zero Trust Architecture
An architectural model in which no network zone is trusted: every request, including one that arrives from inside, is authenticated, authorised and encrypted. NIST SP 800-207 set out the model in 2020, and it is about identity, policy and visibility — not a product list. Why it matters: a bank's or an exchange's internal network is not a security perimeter; an attacker who gets that far is already standing next to the command path, and an internal firewall will not stop them. Read more →
Service Mesh sidecar proxy data plane
An infrastructure layer that manages traffic between services outside the business code: mTLS, service-to-service authorisation, retries, load balancing and telemetry — through a proxy beside each pod or through eBPF at the node. Istio and Linkerd are the common implementations. Why it matters: issuing and renewing certificates by hand in every service breaks on the first update; the mesh gives uniform enforcement of strict mode and one place where you can see who talks to whom. Read more →
SPIFFE and SPIRE Secure Production Identity Framework for Everyone
SPIFFE is a standard for service identity expressed as a URI and issued as an SVID; SPIRE is the implementation that issues and refreshes that identity in practice. The identity is derived from workload attributes — service account, namespace, node — not from an IP address or a DNS name. Why it matters: in Kubernetes pods are replaced daily and addresses change, so location-based identity does not hold; an SVID lets policy and the mesh layer keep working even when the pod is brand new. Read more →
PKI Public Key Infrastructure
The system that issues, distributes and revokes certificates: certificate authorities, a policy document, the trust store, the chain of trust and revocation mechanisms such as CRL and OCSP. Why it matters: every mTLS deployment rests on PKI, and when it is not maintained — certificates that expire, a revocation list that is never refreshed, a root left valid for years — mTLS becomes warning noise in the browser rather than a control. Read more →
CA Certificate Authority
The party that issues certificates and signs them with its private key; it is the root of trust for the whole chain. In an organisation it is usual to keep the root offline and use it to issue intermediates that sign the service certificates. Why it matters: whoever controls the CA controls the identity of every service in the system, which is why its key lives in an HSM and not on a server; a leak of it is equivalent to leaking every certificate at once. Read more →
WAF Web Application Firewall
A component that inspects HTTP requests against signatures and policy and blocks known patterns — SQL injection, XSS, scanning attempts. It sits in front of the application rather than inside it, so it does not know the application's authorisation model. Why it matters: a WAF is a good input control, but it is not an authorisation control; a BOLA request looks perfectly normal to it because it is legal in syntax and signature. Whoever relies on it instead of on an authorisation check finds out from the customer. Read more →
X.509 ITU-T public key certificate standard
The standard that defines the structure of the digital certificate: subject, Subject Alternative Name, validity, key usage, extensions and the issuer's signature. Why it matters: most mTLS failures are born in these fields and not in the code — a missing SAN, an EKU that does not fit the use, a validity that expired in the middle of the night. Whoever reads the certificate before hunting for a bug saves hours. Read more →

Keys & cryptography

HSM Hardware Security Module
Dedicated hardware that generates keys, stores them and performs cryptographic operations inside a physical boundary the key does not leave. The key is born inside and stays inside; the application sends a request and receives a result. Why it matters: a root key in a file or an environment variable, with no root hardware holding it, is a key that has already been stolen; FIPS 140-3 at level 3 requires resistance to physical tampering and identity authentication, and that is what an auditor looks for. Read more →
KMS Key Management Service
A service for managing the lifecycle of keys: issuance, authorisation policy, rotation, destruction and logging of every use. Behind it there is usually an HSM, and the key itself is not exposed to the caller. Why it matters: it separates whoever is allowed to use a key from whoever manages it, and produces an audit trail for every encryption and decryption. Without that separation, permission to read a configuration file turns into permission to decrypt. Read more →
FIPS 140-3 Federal Information Processing Standard 140-3
A US standard for cryptographic modules at four security levels: level 1 requires the use of approved algorithms with no physical requirements, and level 3 adds resistance to physical tampering and identity-based authentication. Why it matters: this is what an organisation is expected to show when it has to prove that its keys are protected; a vendor's marketing statement is no substitute for a validation that is in force, which is why you read the certificate's validity date and not the headline. Read more →
PKCS#11 Public-Key Cryptography Standards #11
A C-language API for talking to a cryptographic module: slots, tokens, sessions and objects such as keys. It exists so that the same application can work against different HSM vendors without a code change. Why it matters: it lets you swap hardware without rewriting the system, but it is also an attack surface of its own — session handling, logout and PIN codes are part of the security policy, not a technical detail for someone else to deal with. Read more →
KEK Key Encryption Key
A key that encrypts other keys rather than data. It sits in an HSM or a KMS, is not written to disk as clear text, and is used to unwrap the layer of keys beneath it. Why it matters: this hierarchy is what allows a root key to be rotated without re-encrypting the entire database; without it every rotation becomes a project of weeks with a defined failure window. Read more →
DEK Data Encryption Key
The key that encrypts the data itself — a customer record, a transaction, a file — and is stored wrapped by a KEK. Why it matters: encrypting each record under a separate key limits the damage of a leak and allows cryptographic deletion: destroying the key disables the information without touching every copy and backup. It is also what lets you hold one KEK and rotate it separately. Read more →
Key Rotation scheduled re-keying
A planned replacement of a key in use with a new one, with a definition of who triggers it, how often, and what happens to data written earlier. Proper rotation requires a version or key id in every encrypted record. Why it matters: a key that never rotates is a key that will be exposed at an unknown date, and as long as there are no versions you cannot know what was encrypted under what; rotation without versions breaks historical decryption. Read more →
PQC Post-Quantum Cryptography
A family of algorithms designed to withstand a quantum computer. NIST published the standards in August 2024 — FIPS 203, FIPS 204 and FIPS 205 — and the draft NIST IR 8547 from 2024 describes the guidance for migration. Why it matters: "harvest now, decrypt later" — encrypted traffic captured today will be decrypted when the hardware is available, and financial data is kept for years. Whoever maps where the algorithms sit today gets ahead of the pain. Read more →
ML-KEM Module-Lattice-Based Key-Encapsulation Mechanism
A lattice-based key exchange algorithm, derived from Kyber and defined in FIPS 203. It is meant to replace the classic key exchange mechanism in TLS and in other interfaces. Why it matters: it enters the connections themselves, so the move to it touches every client and not only certificates; a hybrid combination with X25519 lets you run it without trusting only a new scheme that nobody has tried under real load. Read more →

Trading infrastructure

FIX Financial Information eXchange
A text protocol for passing orders and reports between participants and an exchange, built from tag=value pairs: 35=D is a new order and 35=8 is an execution report. Why it matters: it is the protocol the whole legacy estate around an exchange speaks, so it stays even when something faster exists; but it is text, every field is parsed again, and its latency is measured against binary encoding. Read more →
SBE Simple Binary Encoding
A binary encoding for orders and market messages: every field has an offset and a type fixed in advance, and there is no text parsing in each message. Why it matters: at millions of messages a second every byte and every parse operation counts; SBE cuts decode time and bandwidth, and that is the difference between an engine that holds a load and one that starts to lag behind the market. Read more →
Matching Engine price–time priority matcher
The component that receives orders and matches buyers against sellers by price and then by time of arrival; it usually runs in memory and keeps a deterministic processing order for each instrument. Why it matters: this is where it is decided who got the fill, so any change to the processing order is a change to the rules of the game; it is also the component most worth attacking, because an attack on it moves prices rather than merely taking a service down. Read more →
Order Book resting orders by price and time
The data structure that holds every open order for each instrument, ordered by price and then by time, from which best bid and best ask are derived. Why it matters: it is the real state of the market, and every snapshot sent to the community rests on it; after a failure it has to be rebuilt with absolute accuracy, otherwise the exchange publishes a market that does not exist and nobody knows. Read more →
p99 Latency 99th percentile tail latency
The time from a request arriving to the response leaving, at the 99th percentile: 99% of requests were faster than that value. It is a measure of the tail of the distribution, not of its centre. Why it matters: an average hides exactly what gets in the way; a wide gap between the median and p99 is the difference between a client that gets a price and one that misses a trade, and usually also the difference between normal load and an incident. Read more →
Event append-only journal record
A record appended to a journal that does not change, describing a state change that happened: a new order, a cancellation, a fill or a price change, with an identifier and a monotonic timestamp. Why it matters: when the event is the source of truth you can rebuild state, run controls in real time and analyse afterwards without asking the matching engine; it is also why the data pipeline has to be separate from the order path. Read more →

Regulation & compliance

PCI DSS Payment Card Industry Data Security Standard
A PCI SSC standard for protecting payment card data; the current version is 4.0.1, and the documents are available in the standard's library. It sets requirements for the cardholder data environment — encryption, network segmentation, monitoring and testing — and requires periodic assessment. Why it matters: it is a minimum for anyone touching card acquiring, but it sets requirements rather than an architecture: an organisation can pass an assessment and still run an API that returns someone else's account. Read more →
DORA Digital Operational Resilience Act
An EU regulation for ICT risk management and the operational resilience of financial institutions: mapping dependencies on suppliers, incident management, resilience testing and reporting. Why it matters: it addresses the supply chain and not only the institution itself, so a Singapore supplier serving a European institution is asked to show the same data — dependency mapping, a continuity plan and incident records — even when it is not directly in scope. Read more →
MAS Notice 655 Cyber Hygiene
A notice from MAS on cyber hygiene for financial institutions in Singapore. It sets a baseline — patching, secure configuration, control of administrative accounts and monitoring of access — rather than a list of products or suppliers, and it sits alongside the MAS Technology Risk Management Guidelines, which cover governance, accountability and business continuity. Why it matters: it is the reference point against which a bank in Singapore explains its cyber programme in an audit; translating it into controls happens inside the organisation, so every new system has to enter the programme rather than stay outside it. Read more →
PDPA 2012 Personal Data Protection Act
The Personal Data Protection Act 2012, administered by the PDPC — Personal Data Protection Commission, sets the baseline for how organisations collect, use and disclose personal data. It requires each organisation to appoint a Data Protection Officer, to limit collection to notified purposes, and to notify the PDPC of a notifiable data breach within three calendar days. Why it matters: a security incident stops being internal damage and becomes a reportable event with deadlines and documents; logs, containment and a record of decisions therefore have to exist before the incident, not be built during it. Read more →
BOLA Broken Object Level Authorization
An authorisation vulnerability at object level: the API verifies that the user is signed in, but does not check that the user is entitled to the requested object — an account, a transaction, a file. The OWASP API Security Top 10, 2023 edition, ranks it as the first item. Why it matters: it does not look like an attack because the request is entirely legal; every path that takes an identifier, such as GET /accounts/1234, is a vector, and an authorisation check has to run on every request and not only at sign-in. Read more →
Idempotency idempotency key semantics
A property of an operation whereby running it again with the same identifier has the effect of a single run: the server stores the result and returns it a second time instead of creating another charge. Why it matters: the network drops after the order has already been accepted, and the client sends it again; without an idempotency key you get a duplicate charge that needs a refund, an investigation and an explanation to the customer, and an automatic retry becomes a gamble. Read more →

If you run into a term that is missing here — a new one, another accepted translation, or a definition that looks imprecise — write to me through the contact page. I update the glossary on reader feedback, and every entry added gets a link to the article that goes deeper.

Technical definitions, not legal advice

The definitions here are meant to help engineers read a regulatory document or ask a precise question in a review — not to replace legal advice or a reading of the binding text. The binding text is always the one the regulator published, and in any decision on a control, a report or compliance with a requirement you should work with your legal counsel and the organisation's compliance team.