Skip to content
Bastyx
Book a Demo
Security

Security shouldn’t require blind trust.

How Bastyx works, what it stores, what it never sees, and how to report a vulnerability.

Principles

  • Standards over invention. Authentication is built on FIDO2 and WebAuthn, not a proprietary protocol.
  • No shared secrets to steal. Bastyx removes passwords and long-lived SSH keys rather than storing them more carefully. Where a legacy app can’t drop its password, no person knows it and it’s rotated automatically.
  • Explainable decisions. Every access decision is traceable to the signals it used and the policy rule that applied.
  • Least data necessary. Bastyx never needs biometric data, private keys or the passwords people know, so it never receives them.

Architecture

Bastyx sits between people and the resources they access. Authenticators prove who someone is; Bastyx decides whether that person, on that device, may reach that resource right now, and records the outcome.

  1. EmployeeAlex Morgan
  2. IdentityOne identity, everywhere
  3. DeviceEnrolled and known
  4. AuthenticatorPasskey, key or fingerprint
  5. Bastyx policyAllow, step up or deny
  6. ResourceLaptop, SSH, cloud, app
  7. Audit eventWho, what, where, when

The components are:

  • Directory. People, teams and roles, either managed in Bastyx or synchronized from your identity provider.
  • Device enrollment. Laptops and workstations are enrolled and bound to a person, so policy can require a known device.
  • Authenticator registry. Public keys and credential identifiers for each person’s passkeys and security keys. Private keys stay on the authenticator.
  • Policy engine. Evaluates explicit rules against identity, device, authenticator, resource and risk signals. Returns allow, step up, require approval or deny.
  • Access integrations. Workstation login (macOS, Windows, Linux), SSH through a small agent on each server that trusts the Bastyx user CA, and application sign-in, each asking the policy engine before granting access.
  • Legacy-app access. An identity-aware proxy for internal web apps without SSO, and a vault for passwords that apps can’t drop: random, rotated automatically and never shown to people. Both run next to the apps they protect.
  • Agent identities. AI agents get their own identities, named owners, policies and short-lived credentials, separate from the people who run them.
  • Audit log. An append-only record of authentication events and administrative changes.

Authentication model

Bastyx uses FIDO2: the W3C WebAuthn API together with the FIDO Alliance’s CTAP2 protocol for external authenticators. These are the standards behind passkeys and modern security keys.

How a FIDO2 sign-in works

  1. At registration, the authenticator generates a key pair for Bastyx. The public key is stored by Bastyx; the private key never leaves the authenticator.
  2. At sign-in, Bastyx sends a fresh random challenge.
  3. The authenticator verifies the person locally (a touch, a PIN, or a fingerprint matched on the device) and signs the challenge together with the origin it was requested from.
  4. Bastyx verifies the signature with the stored public key.

Because the signature is bound to the origin, a credential registered for Bastyx can’t be replayed on a look-alike phishing domain. Because nothing secret is stored server-side, there’s no password database to leak.

Phishing and AI-driven attacks

AI makes phishing emails, cloned voices and automated login attempts cheap. None of them help against FIDO2: there’s no secret for a person to reveal, a signature needs a touch or an on-device biometric check, and it’s only valid for the real origin. The remaining target is account recovery, which is why recovery requires admin approval and a second registered authenticator.

Controls on top of FIDO2

  • Policies can require a specific authenticator type, such as a hardware security key, for sensitive resources.
  • Step-up verification when risk is elevated or a resource demands it.
  • Passkey confirmation before an email or AI chatbot prompt that the risk model flags is sent. The model only rates risk; it can never approve anything, so the decision always ends in FIDO2 authentication. High-risk actions
  • Short, policy-defined session lifetimes for infrastructure access.
  • Recovery that requires admin approval and a second registered authenticator, so a lost key can never be used to recover itself.

Data handling

What Bastyx processes:

  • Identity attributes: name, work email, team and role.
  • Device identifiers and the signals policies use, such as enrollment state.
  • Public keys and credential identifiers for registered authenticators.
  • AI agent identities, their owners and the policies that scope them.
  • Outgoing emails and AI chatbot prompts covered by a high-risk action policy: evaluated when sent, then discarded. Only the verdict is kept: who, when, the channel, the categories found and the decision. Content is never used to train models.
  • Vaulted credentials for legacy apps that can’t drop their password: generated by Bastyx, rotated automatically and never shown to people.
  • Authentication and administrative events, including time, resource, device, result and network information such as IP address.

What Bastyx never receives:

  • Passwords people know. There aren’t any in the flows Bastyx manages.
  • Biometric data. Fingerprint and face matching happen on the device, security key or fingerprint reader.
  • Private keys from passkeys or security keys.

Revocation and outages

Revoking a person refuses new SSH connections and closes open SSH sessions immediately, refuses the next laptop unlock, deregisters their authenticators and pauses the AI agents they own. Sessions already open in SAML or OIDC apps last until the app’s own timeout.

If Bastyx is unreachable, open sessions keep running and new SSH and app sign-ins wait. An online laptop that can’t reach Bastyx unlocks with a cached credential; an offline laptop can’t be unlocked. Each protected resource can keep a break-glass account.

Full detail, per access path

Deployment options

  • Bastyx Cloud. Hosted and operated by us.
  • On-premises (customer-hosted). The control plane runs in your own data center or cloud account for teams whose policy or regulation requires it. Legacy-app proxies and vaults run next to the apps they protect in either option.

Responsible disclosure

If you believe you’ve found a security vulnerability in Bastyx or this website, please email security@bastyx.com. Our machine-readable contact details are in security.txt.

Please include

  • A description of the issue and its potential impact.
  • Steps to reproduce, or a proof of concept.
  • Affected URLs, components or versions.
  • How you’d like to be credited, if at all.

While testing, please

  • Only access data that belongs to you, and stop if you encounter anyone else’s.
  • Avoid degrading service for others. No denial-of-service or volumetric testing.
  • Don’t use social engineering or physical attacks against our people or premises.
  • Give us reasonable time to fix an issue before disclosing it publicly.

What we’ll do

  • Acknowledge your report and keep you informed as we investigate.
  • Tell you when the issue is fixed.
  • Credit you publicly if you want to be credited.

Security reviews

If your team needs specific evidence for a security review or vendor questionnaire, tell us what you need and we’ll answer directly.

Security contact

Security reports and questions: security@bastyx.com

Want to go deeper than this page?

Book a demo and walk through the architecture with the engineers who build Bastyx. Bring your hardest questions.