Notes

  • Application Detection and Response - new term like SAST, DAST, SCA
  • We have EDR on local networks
  • We have Wiz of behavior based analysis on cloud
  • We have no real method to secure apps yet, one of the hardest problems to solve
    • Too many languages
    • Hard to find a silver bullet
  • SAST is sometimes very inaccurate, most of time used to triage results
  • We write specific Regex patterns in SAST
  • Similar with WAF, bunch of patterns, hard to predict type of attack to happen
    • Good at finding DDOS attacks, bottleneck in network
    • Hard to use WAF to find application attacks, they are more sophisticated
  • Main idea: no scanning, real time monitoring
  • Instead of: anytime CI/CD runs, static scan, dast scan 8hrs
  • We allow app to run and watch at all times
  • ADR tool will automatically respond
  • We must use an agent to monitor
  • One method is to drop the agent directly on the app server
  • Instrumentation: good for performance
    • Datadog uses these agents for observability, looks at how much percentage of this data is used
      • High # of joins
  • May choose instead to just use eBPF filters to sandbox it, and use OpenTelemetry to give into a hook of the application
    • Less observability
    • Super easy install
  • Agents are watching:
    • Data flow of application
    • Identify if dataflow is missing any security controls
    • Example:
      • Search button, can search and get results
      • Agent can identify as we go through backend, are we applying the appropriate security control
      • If Java:
        • Have you use auto escaping
      • If .NET:
        • Have you used procedures correctly?
    • If untrusted input can cause dangerous function s.t it alters the semantics of the fn, altering logic, then we have a vulnerability 100%
  • Example for SQLI:
    • Someone enters input something AND sleep 5
    • Then, somebody has altered the semantics, easy detection
  • Type Confusion, it can detect this
  • Attacker’s path goes through a different path, different from baseline
  • Contrast Security
  • Agent might introduce additional points of failure
    • Agent follows Hippocratic Oath
    • App runs, but crashes with agent, its fine just don’t attach the agent
    • Performance is really bad, agent should exit out
  • Can’t do this well with AWS Lambda, since these run for 2 or 3s, use SAST instead
    • Lambda not supported well anyways for security
  • Automatically get vulnerabilities fixed with AI Agent
  • Appsec people have no reason to talk to the SOC using Splunk, AWS Pagerduty, but not a lot of logs from appsec
  • If you have ADR, you can feed it into the SOC
  • Agent may have false positives, sometimes apps are insecure by design
  • Site that has two backends, one backend generates HTML and embedded JS, send to the other backend Expression Language Injection
  • Context:
  • AI patches it afterwards given the path is anomalous