Skip to content

PATTERN Cited by 1 source

Platform security at the application layer

Shape

Apply infrastructure-security techniques — continuous monitoring, sampling-based detection, centralized telemetry pipelines, rate-limited inspectors — to the application layer instead of restricting them to the network / OS / proxy layer. Treat application surfaces as infrastructure to be observed.

Named directly in Figma's post (Source: sources/2026-04-21-figma-visibility-at-scale-sensitive-data-exposure):

"We approached this problem with a platform-security mindset — treating our application surfaces like infrastructure and layering continuous monitoring and detection controls on top. By applying techniques typically reserved for lower-level systems to our application layer, we were able to gain continuous visibility into how data moves through our products, without slowing development."

Why it's a mindset shift

Historically:

  • InfraSec monitors flows, packets, syscalls, file-integrity — continuous, sampled, centralized. Think Cilium, Datadog CWP, eBPF kernel hooks.
  • AppSec reviews code, writes unit tests, runs SAST/DAST, coordinates pentests + bug bounty — predominantly ahead-of-time / point-in-time checks.

The mindset shift: take the InfraSec continuous observability playbook and run it inside the app server, on its data flows.

Techniques borrowed from InfraSec

InfraSec technique App-layer analogue
Network-flow sampling Response-body sampling
File-integrity monitoring Schema-drift / unexpected-field detection
Centralized SIEM Unified analytics warehouse for findings
Anomaly detection on traffic Anomaly detection on response shape
Policy as code (firewall rules) concepts/data-classification-tagging
Rate-limited packet inspection Rate-limited response inspection
Two-environment defense (DMZ + internal) Staging + production detection layers

Concrete properties

  1. Continuous, not point-in-time. Runs on every production request (at sampled fraction), not just in CI.
  2. Centralized pipeline across services. Multiple application services post into the same detection backbone — unified schema, unified warehouse, unified triage. Figma's LiveGraph + monolith pattern.
  3. Non-blocking on the hot path. Detection errors degrade detection, never the product (patterns/async-middleware-inspection).
  4. Rate-limited. The detector itself can't be a DoS vector.
  5. Config-driven allowlists for intentional safe-exposure cases (patterns/dynamic-allowlist-for-safe-exposure) so alerts stay signal-dense.
  6. Security engineers ship code alongside product teams — from the post: "Our security engineers ship code alongside product and platform teams, bringing the same creativity and rigor to detection systems that we do to user-facing features."

When this beats adding another preventive gate

  • Your preventive controls are already strong — adding another synchronous gate yields diminishing returns + latency cost.
  • Your product + infra complexity is growing faster than review / test-coverage capacity.
  • You need to catch regressions in already-shipped code, not just block new mistakes.
  • You want to quantify what actually happens in production vs what the threat model assumed.

When prevention still wins

  • Stakes are existential on a single miss (root credentials, payment execution) — detection after-the-fact is too late.
  • The action is destructive + irreversible.
  • The rule is cheap to express as a preventive gate (a type system, a required middleware, a DB CHECK constraint).

Prevention and detection compose; this pattern adds the detection leg to existing prevention, not a replacement.

Adoption hurdles

  • Org structure: AppSec teams often don't own runtime pipelines. Detection-in-depth requires them to ship code in the app-server repo, own on-call triage, and operate data pipelines.
  • FP rate management: early detection layers emit noise; teams that don't invest in dynamic allowlisting + triage workflows lose trust in the alerts fast.
  • Performance discipline: tuning sampling rates + keeping async correctness takes engineering taste the same way p99 tail reduction does.

Example systems

Seen in

Last updated · 200 distilled / 1,178 read