← [ BACK_TO_NOTES ]

Suspicious Login & Risk-Based Authentication

Risk-Based Authentication 用账号、设备、网络、时间、行为与已知泄漏等信号评估登录风险,并动态选择 allow、step-up、delay、deny 或告警;这些信号是概率性上下文,不应替代真正认证因素,也必须避免硬编码单一 IP/设备指纹为身份。

#synthesis #evergreen #auth #observability

[!info] related notes

Suspicious Login & Risk-Based Authentication

一句话关系

Risk-Based Authentication(RBA)不是另一种 authenticator。

它是:

authentication attempt

  • contextual signals → risk decision → choose required control

典型输出:

  • allow;
  • allow but monitor;
  • require MFA / WebAuthn;
  • require reauthentication;
  • delay / CAPTCHA;
  • deny;
  • trigger account protection。

Signal ≠ Proof

常见 risk signal:

  • new device/browser;
  • IP reputation;
  • ASN / proxy / Tor;
  • coarse geolocation;
  • impossible travel;
  • unusual time;
  • many accounts touched;
  • many sources touching one account;
  • credential-stuffing pattern;
  • known breached password;
  • device fingerprint;
  • prior trusted-device history。

这些都不是单独的身份凭证。

例如:

same IP ≠ same user

same fingerprint ≠ same physical device

new IP ≠ attacker

New Device

“New Device” 本身通常是:

no matching prior client/session/trust context

而不是:

cryptographically proven new hardware

浏览器升级、清 Cookie、换 Profile、隐私模式都可能产生“新设备”。

所以新设备更适合:

risk score increase → maybe step-up

而不是:

new device → hard deny

Impossible Travel

如果用户:

09:00 Tokyo 09:20 London

可能是:

  • VPN;
  • proxy;
  • mobile carrier;
  • IP geolocation error;
  • shared corporate egress;
  • actual account compromise。

因此 impossible travel 是 investigation/step-up signal,不应机械等同攻击事实。

Velocity / Graph Signals

真正有效的风控往往是关系特征:

one source → many accounts

one account → many sources

one device → multiple unrelated accounts

valid password → repeated MFA failure

new device → immediately changes email/MFA

这些通常比单个 IP 黑名单更有解释力。

Decision Policy

一个简化状态机:

low risk → normal authentication

medium risk → additional MFA / WebAuthn → reduced trust → notification

high risk → deny or restricted recovery → revoke suspicious session → alert user/security team

具体阈值必须可配置、可审计。

Step-Up 而不是直接封禁

风险信号有误报。

因此很多场景更适合:

risk elevated → ask for stronger proof

而不是:

risk elevated → permanent lock

这能降低:

  • false positive;
  • travel/VPN friction;
  • attacker-triggered DoS。

Known Device / Trusted Device

历史设备可以降低风险分数,但:

trusted device ≠ authentication factor by itself

如果 trusted-device token 被复制,攻击者也可能继承低风险路径。

所以仍需要:

  • revocable trust credential;
  • expiry;
  • anomaly detection;
  • high-risk operation reauth。

详见 Trusted Device & Remember MFA。

Risk 与 Reauthentication

风险事件可以让当前 Session:

assurance downgraded

然后:

ordinary read → continue

sensitive action → require recent strong reauthentication

详见 Authentication Assurance & Reauthentication Freshness。

Notification

高价值 suspicious-login event 可以通知用户:

  • new login;
  • rough time/device/location;
  • action if not recognized;
  • revoke sessions / secure account link。

不要在通知里暴露:

  • internal risk score;
  • detection rule;
  • secret/token。

Privacy

RBA 很容易扩张成过度追踪。

Device fingerprint、location、behavioral biometrics 都可能带来隐私和合规成本。

需要:

  • data minimization;
  • purpose limitation;
  • retention;
  • access control;
  • transparency where appropriate。

Adversarial Adaptation

攻击者会适应规则:

IP block → residential proxy

UA check → real browser

fixed threshold → low-and-slow

所以风控系统应:

  • 多维信号;
  • 可更新规则;
  • monitoring;
  • feedback loop;
  • 不把单一 heuristic 当安全边界。

学习检查

  1. Risk-Based Authentication 为什么不是新的认证因素?
  2. same IP / same fingerprint 为什么不能证明 same user/device?
  3. new device 为什么更适合 step-up,而不是默认 hard deny?
  4. impossible travel 有哪些常见 false positive 来源?
  5. velocity / graph signal 为什么比单一 IP 更有价值?
  6. suspicious event 怎样让 Session assurance 降级?
  7. Trusted Device 与 risk scoring 应怎样配合?
  8. 设计一个 medium-risk login 的 allow/step-up/notification 流程。
创建于 2026/9/30 更新于 2026/9/30
Digital Biome
Access

键盘导航

j / k / ↑ / ↓
选择对象
Enter / o
打开选择
/ 或 Ctrl+K
全站搜索
i / f / 1–4
筛选当前列表
g → h / l / p / t / a
首页 / 笔记 / 项目 / 工具 / About
r → 1–9
打开所选对象的关联
u
返回上一页
Esc
关闭浮层 / 取消命令;输入框内清空文本