Threat protection
Login and session defenses: bot detection, per-IP limits, brute-force lockout, breached passwords, impossible travel.
Threat protection guards the login, signup, and password-reset surfaces against automated abuse and compromised credentials. These controls run on every relevant request, independently of authentication policies, and several are on out of the box.
Configure them under Security → Threat protection, where each control has its own tab. The landing page lists all five controls, each with a live Enabled / Disabled status indicator so you can read your posture at a glance.
Note: Breached-password detection and the bot detection (response Block for login, signup and password reset) are enabled by default on a new instance. An instance set up earlier keeps its configured values; its former CAPTCHA settings became these three responses (Never and When risky, which showed no CAPTCHA in practice, became Monitor, Always became Block). The other controls are available immediately and recommended.
The five layers
Each control defends a different target. Layer them rather than relying on any one.
| Target | Control | What it stops |
|---|---|---|
| The source | Per-IP rate limiting | One address working through many accounts: attempts per IP are capped in a time window. |
| The account | Brute-force protection | Repeated failures against one account: further attempts are stopped after a set number. |
| The credential | Breached-password detection | Passwords that already appear in public breaches, refused wherever a password is set or presented. |
| The request | Bot detection | Scripted traffic on the login, signup, and reset forms; suspicious requests solve an invisible computation, clear bots are rejected. |
| The session | Impossible-travel detection | A live session reappearing from a place it could not have reached in the elapsed time, a sign of a stolen session cookie. |
Per-IP rate limiting
The Per-IP rate limiting tab caps how many login or signup attempts one IP address may make in a window, which slows credential stuffing and account enumeration:
- Login attempts are counted per day per IP; exceeding the threshold blocks that IP for a day.
- Signup attempts are counted per minute per IP; exceeding the threshold blocks that IP for an hour.
You can keep the recommended defaults or set custom thresholds, and you can exempt trusted network zones. Custom thresholds that equal the defaults are saved as the default setting.
Brute-force protection
Where per-IP rate limiting counts attempts per address, brute-force protection counts consecutive failed logins for a specific account from a specific IP. After the configured number of failures it stops further attempts for a time. You choose:
- The number of consecutive failures allowed.
- What happens once it is reached: further attempts from that address are stopped, the account is put into the locked state, or both.
- Whether the account owner receives an email when the account is locked.
Trusted network zones are exempt. A successful login resets the counter.
Where do I undo a block? This is the most common support question, and the answer depends on what tripped:
- A brute-force account lockout is lifted by unlocking the user in Directories → People (the user can also clear it with a password reset).
- An IP block (from this control or from per-IP rate limiting) is lifted in Audit Trail → Rate Limits.
See rate limits & blocked IPs for the IP side and user lifecycle for the account side.
Breached-password detection
The platform checks credentials against a database of passwords known to have appeared in public breaches (a privacy-preserving lookup: only a short hash prefix leaves the system, never the password). The check runs at every point a password is set or presented, and you decide per moment how it reacts:
| Moment | Reaction |
|---|---|
| A password is chosen at signup, on a password change, or on a reset | The breached choice is rejected at the form, so the user picks a safe one (Block new credentials). |
| A password that leaked after it was chosen is used to sign in | The sign-in is refused and the user must reset the password before continuing (Block account). |
| A user's password turns up in a breach | The user receives a warning email with a reset link (Inform the user). |
When the check blocks, the user sees a clear message that the password appeared in a known breach and must be changed. Detection runs transparently, and the event is recorded in the audit log.
Pair this control with the per-directory password policy (minimum length, dictionary/breach check, history) as a single defense — the policy sets the bar for new passwords, breach detection catches credentials that were safe when chosen but later leaked. See user directories.
Bot detection
The Bot detection tab replaces the former CAPTCHA. Every request to the password login, the emailed-code login, the signup and the password-reset form is scored from signals the platform observes itself; there is no image to decipher and no external reputation service. The result is one of three verdicts: allow, challenge (the request is suspicious) or block (a clear bot). You choose how the platform responds in three settings: Response on the login (the password form and the emailed-code form share it), Response on the signup and Response on the password reset:
- Monitor only counts. Users notice nothing; the dashboard shows which signals fired and what Block would have done.
- Block acts on the verdict. A suspicious login has to solve a small computation that the page performs in the background before the form is accepted (one to a few seconds, no click, no typing); the signup and password-reset forms always carry this computation, started while the page is open, so a person never waits for it; a clear bot is rejected after a short delay. A browser that has signed in to the same account before lowers the rating; requests from a trusted network zone are never blocked.
New instances start with Block in all three settings. Start with Monitor when unsure and switch to Block once the dashboard shows few false alarms; the security health report reminds you when all three only count. The switch at the top turns the detection off entirely; nothing is scored or counted then.
Principles
- Block only on hard evidence. A request is rejected as a clear bot only on evidence that a real browser cannot physically produce. Soft signals, which hit people regularly, and too many attempts never block on their own; at most they make a request suspicious.
- An invisible computation instead of a puzzle. A suspicious request solves a small computation that the page performs in the background; there is no image to decipher, nothing to click and nothing to type.
- A client is address plus browser. For the login, a client is the combination of address and browser, never the address alone.
- Signals describe the client, never the account. The result is the same whether or not the entered account exists, so the detection does not reveal which accounts exist.
- No third party. The platform observes every signal itself; no external reputation service is involved.
What is recorded
Every suspicious or blocking verdict is kept for 30 days together with the signals that fired, and false alarms are measured: when the same browser goes on to sign in, the verdict counts as a confirmed person. The dashboard's Bot detection signals table shows, per signal, how often it fired, how often it hit a person and how often the verdict was block. A blocking verdict also counts as Bot activity in the attack chart (blocked with Block, detected only with Monitor) and writes a PRE_LOGIN_BOT_DETECTION entry to the audit log.
The signal reference, with every signal and how the signals combine into a verdict, is available to signed-in users under Bot detection signals.
Trusted-network bypass
Each of the above controls accepts a list of trusted network zones to exempt. Traffic from a trusted zone, your corporate network for example, skips the bot detection and rate limiting. Define these zones first under network zones, then select them on each protection tab.
A trusted zone is meant for traffic you already vouch for — your corporate egress or a company VPN — so high-volume office traffic is not falsely throttled or challenged. Use this sparingly: an exempt zone removes a layer of defense for anyone inside it.
Impossible-travel detection
A session is acted on when a request arrives from a location that could not be physically reached from the session's last recorded location in the elapsed time (a travel speed above roughly airplane speed). This catches a session that continues from a second place at the same time, a common sign of a stolen session cookie. The response is configurable (see below) and the event is recorded in the audit log.
The check runs while a session is active and is on by default. It needs geolocation enabled and never triggers when a location cannot be resolved; requests from Tor, iCloud Private Relay, or Starlink are also ignored, because their location is unreliable and shifts legitimately. Only physically impossible jumps trigger it, so a normal trip, even across borders, does not end the session, and a short move within the same region is never treated as travel.
Configure it under Security → Threat protection → Impossible travel:
| Setting | What it does |
|---|---|
| Enabled | Turns the check on or off. On by default. |
| Response | What happens on detection. Require re-verification: the user is sent through a fresh MFA re-challenge and keeps the session (users without an enrolled factor re-enter their password); the least disruptive option for legitimate travellers, and a re-verification rather than hard containment. Sign out of this session (default): end this session and require a new sign-in. Sign out of all connected apps (Single Logout): also sign the user out of every connected application via OIDC back-channel logout and revoke issued tokens (best-effort, reaches applications that support back-channel logout); appropriate when you treat an impossible move as a likely account takeover. |
| Sign out on any country change | Off by default. When on, any change of country triggers the response regardless of travel speed. Coarser than the velocity check, but a stronger backstop. See when to use it below. |
| IP exclude list | Network zones whose traffic is exempt from the check. |
When to use "sign out on any country change": it is a stronger backstop than the velocity check. It also catches a session that sits idle and then resumes from a distant country, which the velocity check alone cannot flag. The tradeoff is false positives for users who legitimately change countries (near a border, cross-border commuters, mobile roaming); exempt their networks with the exclude list.
Because the check can be tripped by a VPN or a mobile network that routes through a distant gateway, add those egress ranges to the exclude list if legitimate users are signed out unexpectedly.
Security-event emails
The platform emails the affected user when key account-security events happen, so a compromise is noticed even if the admin is not watching:
- A password change.
- A change of the account's email address.
- An MFA method reset.
- An MFA method enrolled.
- A sign-in from a device the platform has not seen before. This one is sent by default and can be made optional or turned off under Settings → Notifications.
Each of these events is also written to the audit log. If a user questions an email they received, you can correlate it with the matching audit entry. The sender address and look of these emails are configured under Branding → Email.
Mapping to MITRE ATT&CK
The dashboard's attack-detection row groups what these controls catch into attack types. Each type corresponds loosely to a technique from the MITRE ATT&CK framework; the table names the closest technique and which control feeds the numbers. The mapping is an orientation aid, not a certified coverage claim.
| Attack type on the dashboard | MITRE ATT&CK technique | Fed by |
|---|---|---|
| Brute force | T1110 (Brute Force) | Brute-force protection throttling and account lockouts, for wrong passwords and wrong MFA codes alike; sustained failed client authentications on the API endpoints count as detected |
| Password spraying | T1110.003 (Password Spraying) | Per-IP rate limiting on failed sign-ins: one address trying many accounts |
| Access using a valid account | T1078 (Valid Accounts) | Breached-password detection and impossible-travel detection |
| Session token theft | T1539 (Steal Web Session Cookie) | Session anomaly detection: a session presented from a different browser is terminated; a mid-session country change that no policy acts on counts as detected |
| Anonymized network | T1090 (Proxy) | Sign-ins from the Tor network, public VPN services, or open proxies, blocked or flagged depending on your policy (requires geolocation) |
| Bot activity | T1110.004 (Credential Stuffing) | Bot detection verdicts of a clear bot on sign-in, signup and password reset (blocked with response Block, detected only with Monitor) |
Phishing does not appear as a type on purpose: the platform prevents phishing with phishing-resistant factors but has no detector for it, and the dashboard only shows what is actually measured.
Related
- Network zones — define the trusted zones used for bypass.
- Authentication policies — add risk-based step-up on top of these baseline defenses.
- Audit logs — review breach, lockout, and rate-limit events.