How the platform protects identities, credentials and data
For the administrators, security officers and decision-makers evaluating an IAM solution, with a technical reference for security reviewers: how the platform keeps access secure, and where its guarantees end.
This document explains how the platform protects identities, credentials, and customer data. It is written for IT administrators, security officers, and decision-makers evaluating an identity and access management (IAM) solution, and closes with a technical reference for security reviewers who want the trust model, the algorithms and the exact behaviour.
The security model rests on a few principles: defend in depth rather than rely on a single control, build on recognised standards (OIDC, SAML, FIDO2/WebAuthn, NIST 800-63) instead of proprietary schemes, collect the minimum data needed, and make every security-relevant decision auditable.
Every statement describes the product as it is, not a plan. Where a property depends on the plan or the deployment, the text says so; where it stops, the section Where the line is drawn says that too.
Security at a glance
These are the capabilities that set the platform apart. Each is explained in more detail in the sections that follow.
Authentication and multi-factor security
Assurance levels you can reason about
Authentication strength follows the NIST 800-63 assurance model (AAL1–AAL3) rather than a simple "MFA on/off" switch. Each combination of factors a user presents resolves to a defined assurance level, and applications or policies can require a minimum level for access. This makes it possible to state precisely how strongly a given session is authenticated, a meaningful distinction for regulated environments and audits.
Weaker factors are treated honestly. Email-based verification, for example, is capped at the lowest assurance level because it is not a true out-of-band authenticator under the standard. It can never be mistaken for strong second-factor authentication.
Phishing-resistant, hardware-backed multi-factor
The platform supports passkeys and security keys (FIDO2/WebAuthn) alongside authenticator apps (TOTP), SMS one-time codes, and one-time recovery codes. The highest assurance level (AAL3) is granted only to hardware-bound security keys whose origin is cryptographically verified against the FIDO Alliance metadata service. Synced or software-based passkeys are recognised but treated as a lower level, so a claim of "hardware-grade authentication" actually corresponds to a non-exportable key.
Multi-factor requirements can be configured separately for your workforce and for external users, so a single deployment can mandate strong MFA for staff while offering a smoother, conditional experience to customers or partners.
Password protection
For accounts in the built-in directory, passwords are stored only as bcrypt hashes (cost factor 12), never in clear text. Password policy can enforce:
- rejection of known-breached passwords, checked through a privacy-preserving lookup in which only a short hash prefix leaves the system,
- rejection of common passwords from a maintained dictionary of the most frequently used credentials,
- a ban on passwords that contain the user's own name or email,
- password history, preventing reuse of recent passwords, and
- periodic expiry.
No lock-outs, no silent gaps
Users protected by mandatory MFA cannot remove their last remaining factor, closing the window where an account would briefly have no second factor. When an administrator initiates an MFA reset, the existing factor remains valid until the user has successfully enrolled a new one. Recovery is handled through a single-use, time-limited link rather than by deleting a user's factors outright, so an account can never be left both locked and unprotected.
Account-security notifications
Users are emailed when their password or email address changes, when a second factor is enrolled or reset, and, by default, when they sign in from a device the platform has not seen before. A compromise is noticed by the person it affects, even when no administrator is watching.
Single sign-on and protocol security
The platform acts as an OpenID Connect (OIDC) provider, a SAML identity provider, and a WS-Federation provider, so applications can integrate over the protocol they already support. Across all of them, signed assertions and tokens give relying applications cryptographic proof of origin.
Hardened OIDC
- Proof Key for Code Exchange (PKCE) is required for public clients such as single-page and mobile apps, eliminating a well-known class of authorization-code interception attacks.
- Sender-constrained tokens bind an access token to the client's own key pair (DPoP) or, on deployments with a mutual-TLS edge, to its client certificate. A token captured in transit or at rest cannot be replayed, because the attacker cannot prove possession of the key.
- Issuer identification is included in authorization responses, protecting deployments that federate to multiple identity providers from mix-up attacks.
- Token exchange and resource indicators let a service request a narrowly scoped token for exactly the API it needs, rather than passing a broad, over-privileged token between systems.
- Refresh tokens rotate and have bounded lifetimes, so a leaked token cannot be reused indefinitely.
Financial-grade profiles for high-assurance APIs
Applications with elevated protection requirements can be placed on a FAPI (Financial-grade API) security profile. The profile is selected per application, so existing integrations are unaffected by enabling the capability. Where it is active, the profile's requirements are enforced rather than merely documented: the application must be a confidential client, must authenticate with its own key or a client certificate instead of a shared secret, is restricted to the authorization-code flow, must push its authorization request to the provider before sending the user to it, receives sender-constrained access tokens, and is pinned to PS256 signatures.
FAPI 2.0 is implemented in full. FAPI 1.0 Advanced provides the profile's protocol behaviour, including signed authorization requests and signed authorization responses. The certificate-bound access tokens the profile also mandates require a dedicated deployment with a mutual-TLS edge. Neither profile has been submitted to the OpenID Foundation conformance suite, so the implementation is described here as profile-aligned rather than certified.
SAML and federation
Incoming SAML requests are validated against the protocol schema before they are processed, which narrows the surface for signature-wrapping and malformed-XML attacks. Federation metadata is signed, and when remote metadata is imported, the source is validated to prevent it from being used to reach internal systems.
Federated and social login
Users can sign in through external identity providers (your own corporate IdP, or social providers) using standards-based federation. Each federated sign-in is protected against cross-site and code-injection attacks through state, nonce, and PKCE validation, and external profiles are mapped onto local accounts in a controlled, idempotent way. An email domain claimed by one directory cannot be taken over by a sign-up or a federated login into another.
Encryption and key management
In transit, all traffic is protected with TLS. Connections use modern TLS versions, including TLS 1.3, with Perfect Forward Secrecy (PFS), so compromising a key later cannot decrypt past sessions, and HTTP Strict Transport Security (HSTS) tells browsers to refuse any downgrade to unencrypted HTTP. Outbound integrations negotiate modern TLS versions as well.
At rest, the credentials that matter most are never stored in a recoverable form at all: passwords and recovery codes are kept only as bcrypt hashes, so even a full copy of the database yields no password. Second-factor seeds, the shared secrets behind authenticator apps (TOTP) and hardware tokens (OATH), are protected by field-level envelope encryption: each value is encrypted with AES-256-GCM under a data key of the instance, and that data key is stored only in wrapped form, under a key-encryption key that belongs to this one instance and never leaves a key management service outside it. The ciphertext is bound to its field and its key, so it cannot be copied into another field and decrypted there. A database dump or a backup therefore carries none of these secrets in usable form. Field encryption is switched on per deployment, and every new deployment starts with it. The secrets it does not cover are named under Where the line is drawn.
Active Directory / LDAP credentials follow a hold-your-own-key (HYOK) model: the key that protects them is generated and held inside your own network, never by the hosted service. When an on-premises directory is connected through the Enterprise Connector, the user's password is encrypted in the browser before the form is submitted, using RSA-OAEP with SHA-256 through the browser's built-in cryptography API. The key pair is generated by the Enterprise Connector itself and registered when the connector is provisioned: the first connector for a directory establishes the public key, and every further connector for that same directory must present an identical key or its provisioning is refused. The private half never leaves your network. The hosted service relays a ciphertext for which it holds no key.
Signing keys used to issue tokens are RSA-3072; a deployment whose keys were created before this key size keeps signing with RSA-2048 until an administrator rotates them. Rotation is gradual: new keys take over for issuing while previously issued tokens remain valid until they naturally expire, so rotation never disrupts active sessions. SAML and WS-Federation applications sign their assertions with their own copy of the key that was current when their certificate was issued; rotation leaves that copy in place, renewing the application's certificate replaces it. Internal secrets, such as the keys that sign session cookies, are derived from a rolling key pool that rotates automatically every 30 days without logging anyone out.
Post-quantum readiness is handled as a matter of crypto-agility rather than premature migration. Every cryptographic primitive in the product is documented in a maintained inventory (effectively a cryptographic bill of materials) together with its exposure to quantum attack, which is what makes a staged migration plannable instead of speculative. The symmetric algorithms and hash functions in use are unaffected by known quantum attacks and need no replacement. Signatures on tokens and assertions only need to hold until they are verified, and those lifetimes are measured in minutes to hours, so they carry no "harvest now, decrypt later" risk; the existing key-rotation machinery is the migration vehicle once the OpenID Connect, SAML and FIDO ecosystems publish interoperable post-quantum algorithms. Priority therefore sits with encrypted material that is recorded or stored long-term: the transport layer is the first target for hybrid key exchange, and the envelope formats carry an explicit version so a post-quantum wrapping can be added without re-encrypting anything by hand. This follows the timeline set out in NIST IR 8547 and the German BSI's recommendation to adopt hybrid schemes.
The complete list of algorithms is in the technical reference.
Tenant isolation and access control
Isolation by architecture, not by query
Many multi-tenant platforms store every customer's data in shared tables, separated only by a tenant identifier on each row, a model where a single missing filter in application code can leak data across customers. This platform takes a different approach: each tenant is a separate deployment with its own database and its own set of nodes. There is no neighbouring tenant's data to reach in the first place. Isolation is a property of the infrastructure, not something that depends on every query being written correctly. The key management service that holds the tenants' encryption keys keeps the same boundary: each instance authenticates with its own identity and can use only its own key.
Role-based access control
Administrative access is governed by a hierarchy of roles, from read-only viewer through to owner, with each role granting a defined set of permissions on each resource. Permissions distinguish between acting on one's own data and acting on others', and access to sensitive information, such as user risk data, is restricted to higher-privilege roles. Administrative sessions require a fresh sign-in after at most eight hours.
For organisations that delegate administration, an organisation administrator can manage users, groups, and roles within their own organisation without being granted system-wide access.
Emergency access without a standing backdoor
If the regular login path fails, whether through a directory outage or a lost owner account, administrative access still has to be recoverable. The usual answer is a shared emergency password that quietly becomes a permanent weak point. This platform solves it differently. A dedicated standby account can be prepared that has no password at all and can only be used with an enrolled hardware security key, over a separate sign-in path. It is considered armed only while such a key is enrolled, so an unconfigured deployment has no emergency entrance to attack.
The standby account holds no stored privileges. Owner rights exist only as a time-boxed grant that is re-evaluated on every request, so access ends when the grant's expiry passes rather than when a cleanup job happens to run, and it can be surrendered at any time from a banner that is visible on every page for the duration. The account is deliberately confined. It cannot be assigned applications, roles, or groups, it cannot obtain tokens for any connected application, it does not appear in user lists or pickers, and it is not counted as a licensed user. Enrolling or removing a key and every emergency sign-in are written to the audit trail and additionally notify the owners by email, so the control cannot be prepared or used unobserved.
Programmatic access
Machine-to-machine and API access is governed by granular scopes: a client receives only the specific permissions it has been granted, and policy controls determine whether access requires an explicit grant. This keeps automated integrations to least privilege.
Device trust and adaptive access
Access policies are not limited to identity. They can take into account the context of each sign-in and, where device management is in use, the health of the device itself.
Conditional access policies can combine conditions such as network zone, time of day, group membership, and required assurance level, and can require re-authentication at defined intervals. Trusted corporate networks can reduce friction, while access from defined high-risk contexts can require stronger authentication or be blocked entirely.
Device posture can be made a condition of access. Managed devices can be required to meet compliance criteria (for example disk encryption enabled, or a firewall active) before access is granted. The login page asks the device itself for a signed statement of enrollment and compliance, bound to the sign-in attempt, so the decision rests on the device's own key rather than on anything the browser claims. The endpoint agent that provides this runs with high privileges on every device; its own trust model, cryptography and command surface are documented separately in the agent's safeguards document.
User risk scoring gives administrators a continuously maintained posture score for every user. The score combines impact (how much access the account has) with likelihood (how exposed it is: missing MFA, a breached password, a dormant account, and similar factors). Because it is computed from posture rather than only from sign-in anomalies, it answers a question administrators actually ask: which of my accounts are most dangerous right now?
Threat protection
The login, sign-up and password-reset surfaces are the most attacked part of any identity platform. Five layers protect them, each aimed at a different target, and they run on every relevant request independently of the access policies above.
| Target | Control | What it stops |
|---|---|---|
| The source | Per-IP rate limiting | One address working through many accounts (password spraying) |
| The account | Brute-force protection | Repeated failures against one account; stops further attempts, locks the account, or both, and can notify the owner |
| The credential | Breached-password detection | Passwords that appear in public breaches, refused when chosen and, if they leak later, at sign-in |
| The request | Bot detection | Scripted traffic and credential stuffing on the login, sign-up and reset forms |
| The session | Session anomaly detection | A session cookie replayed from another browser, or a session that jumps to a place it could not have reached in the elapsed time |
Bot detection
The bot detection replaced the classic CAPTCHA. It separates detection from response: every request is scored, and the administrator decides per form whether the verdict is only recorded (Monitor) or acted on (Block). New instances start with Block on all three forms.
- Signals a script cannot fake. A hard signal is something a real browser cannot produce. Soft signals hit real people regularly and are weighted and capped accordingly. Velocity counts identifiers, attempts and failures per client, where a client is address and browser together, never the address alone.
- Block needs evidence, not just a score. A request is rejected as a bot only on hard evidence; soft signals and velocity alone never block. A person behind a VPN with JavaScript off is suspicious, never blocked.
- An invisible proof of work instead of a puzzle. A suspicious request has to solve a small computation that the browser performs in the background (well under a second to a few seconds, no click, no typing). Sign-up and reset forms always carry it, solved while the page is open. Each challenge is signed, expires after an hour and can be used once. A request judged a clear bot is held for a few seconds and then refused, without its credentials ever being checked.
- No account oracle. Every signal describes the client, never the account, so the verdict is the same whether or not the entered account exists. The bot detection cannot be used to find out which accounts are real.
- No third party, no tracking. The detection runs entirely within the platform. No external reputation service, no challenge provider, no cross-site tracking of your users.
- Measured, not guessed. Every suspicious verdict is kept for 30 days with the signals that fired; when the same browser then completes a sign-in, the verdict is marked as a false alarm. The dashboard shows, per signal, how often it fired and how often it hit a person.
Session protection
A session is bound to the browser it was created in; presented from a different browser, it is terminated. Impossible-travel detection acts when a session continues from a place it could not have reached in the elapsed time, a typical sign of a stolen session cookie. The response is configurable: a fresh multi-factor re-verification, the end of this session, or the end of the session together with a back-channel sign-out of every connected application and revocation of its tokens. Each session also has a defined maximum lifetime and an inactivity timeout.
Seeing attacks, not only stopping them
The administration dashboard shows what the controls above catch, grouped into attack types (brute force, password spraying, valid-account use, session theft, anonymised networks, bot activity) and mapped to the corresponding MITRE ATT&CK techniques, together with a map of attack origins. Administrators can define alarms over the audit trail, such as a run of failed administrator sign-ins or a policy change, and be notified by email when one fires.
Auditing and transparency
Every security-relevant event (sign-ins and failures, multi-factor actions, administrative changes, role and policy changes, device commands, and more) is recorded in a structured audit trail, complete with the actor, source context, and outcome. This gives administrators a clear forensic record of what happened, to whom, and from where.
The audit trail is tamper-evident. Shortly after an entry is written it is sealed: each record is linked to its predecessor by an HMAC-SHA256 hash chain and signed with Ed25519, so removing, reordering, or altering an entry breaks the chain at a detectable point. The verification key is retrievable through the API, and a batch of exported records can be checked against it, which means an archive kept outside the platform can still be proven intact years later.
Sensitive values are automatically redacted before anything is logged: passwords, tokens, keys, and similar secrets never reach the log, on any path. This prevents the common failure mode where credentials leak into log files or error-tracking systems. Message deliveries (mail and SMS) are recorded with metadata only, never with the content.
Sessions that end, whether by expiry, inactivity, or a security event, are recorded with the reason rather than disappearing silently.
Security testing and engineering
Security is verified, not assumed.
- Internal penetration testing probes the running platform the way an attacker would, targeting authentication, access control, and the exposed surface. Weaknesses found this way are remediated and then re-checked.
- AI-assisted source-code auditing goes beyond black-box testing: the entire codebase is reviewed by hand with AI assistance, so coverage spans the whole source rather than a sampled part of it. What the review surfaces feeds the same fix-and-verify loop.
- Security invariants are tested. The server is written in strictly typed TypeScript, and an automated test suite gates every staging and production build. Load-bearing security rules, such as the scoping of workforce and external users or the completeness of the audit-event catalogue, have their own guard tests that fail the build when a change breaks them.
- A maintained cryptographic inventory. A new algorithm or key without an entry in the inventory is treated as a review finding.
- Security fixes take priority. There is one supported version. Security fixes are classified by severity and take an expedited path through the same staged rollout as every release, with a rollback to the last proven version as the first response to a faulty update.
Compliance and data protection
The platform is operated with EU data residency and is designed around the principle of data minimisation: collecting only the identity data needed to deliver the service. It is operated entirely in-house, with no third-party sub-processors: every part of running the service, the mail path included, sits on DTS's own infrastructure, so your identity data is handled only by DTS. Data leaves that boundary only where you send it yourself, through the applications, provisioning targets, identity providers, mail providers and SMS gateways you connect. Those are your own data flows: you set them up and you can switch them off again at any time.
Its controls are documented against the German BSI C5 (Cloud Computing Compliance Criteria Catalogue), mapping product capabilities (authentication, multi-factor, session management, role-based access, password protection, and audit logging) to recognised criteria. This work is maintained alongside the product so that the documented control set reflects what the software actually does. The platform as a cloud service and the software it runs on hold a BSI C5:2026 Type 1 attestation from an independent auditor, which confirms that the controls were suitably designed and implemented as of the reference date, 15 July 2026. The report is confidential; customers and prospects can request it from sales. The operating company, DTS Systeme GmbH, holds an ISO/IEC 27001:2022 certification.
For organisations in scope of NIS2, the platform carries the identity share of the risk-management measures the directive asks for: multi-factor and phishing-resistant authentication, access control on the least-privilege principle, and an audit trail that shows who held which access and when. The obligation falls on the organisation rather than on its software, so the platform enforces those measures and produces the evidence for them; it does not discharge the duty.
Vulnerability reporting and incident response
Reporting a vulnerability
If you believe you have found a security vulnerability, we want to hear about it. Report it to the security contact and we will acknowledge the report, investigate, keep you informed of progress, and fix confirmed issues on a priority track.
We ask for coordinated disclosure: please allow a reasonable window to remediate before any public write-up, and act in good faith, without accessing, altering, or deleting data that is not yours. There is no paid bug-bounty programme today; every report is handled on its merits.
Security contact: incidents@dts.de
The same contact is published in machine-readable form at /.well-known/security.txt.
Incident response
A defined incident-response process governs how security events are detected, contained, and recovered from. If an incident affects your data, we notify you without undue delay through your administrative contact, share what is known and what is being done, and follow up once the full picture is clear. The tamper-evident audit trail described earlier supports the forensic record this depends on.
Technical reference
Who is trusted with what, what a compromised instance can and cannot do, the cryptographic algorithms in use, how the platform behaves when something is missing, and where each protection ends.
The technical reference is available to registered users. Sign up with your company email address.
Open the technical referenceSecurity questions, answered
The questions security officers and compliance teams ask first.
Where does DTS Identity store and process data?
DTS Identity stores and processes data in the European Union. DTS Systeme GmbH operates the service in-house, without third-party sub-processors for identity data, and each customer runs in a dedicated environment with its own database and its own nodes rather than a shared table with a tenant column. Anything you connect yourself, such as an application, a provisioning target or your own mail provider, is your own data flow.
Is DTS Identity GDPR compliant?
DTS Identity is designed around the GDPR: data stays in the EU, only the minimum data needed is collected, and every security-relevant decision is auditable. GDPR alignment is a design principle rather than a certificate; the ISO/IEC 27001:2022 certification is held by the operating company, DTS Systeme GmbH.
How does DTS Identity protect credentials?
DTS Identity stores passwords in the built-in directory only as bcrypt hashes. When an on-premises directory is connected, the password is encrypted in the browser for an Enterprise Connector inside the customer's own network, and only that connector holds the private key. Second-factor seeds are envelope-encrypted at rest under a key held in a key management service outside the database, and every new deployment starts with this field encryption switched on.
Which authentication methods are phishing-resistant?
Passkeys and hardware security keys based on WebAuthn and FIDO2 are the phishing-resistant methods in DTS Identity, because the credential is bound to the origin it was registered for and cannot be replayed against a lookalike site. TOTP, SMS and recovery codes are available as additional factors.
How does DTS Identity stop bots and credential stuffing without a CAPTCHA?
DTS Identity scores each sign-in, sign-up and reset request from signals a real browser cannot fake. Suspicious requests solve an invisible proof of work, clear bots are rejected after a delay, and the result never reveals whether an account exists. No third-party service is involved.
What stops a stolen access token or session from being reused?
DTS Identity can bind access tokens to the client that requested them (DPoP, or mutual TLS on deployments with an mTLS edge), so a token lifted from one client is useless anywhere else. Sessions are bound to their browser and are ended when they reappear from another browser or from a place the user could not have reached.
How do I report a security vulnerability?
Security vulnerabilities in DTS Identity go to incidents@dts.de, the contact also published in machine-readable form at /.well-known/security.txt. Reports are acknowledged, investigated and fixed on a priority track under coordinated disclosure; there is no paid bug-bounty programme today.
Evaluating the platform for your organisation?
Talk to our team about a deployment, a proof of concept, or a deeper security review.