AI 101

Side-channel attacks

Also known as: side channel

Security

A side-channel attack infers secrets from a system’s behavior rather than its output — timing, memory access, power draw.

  • Inference timing can reveal what’s being processed.
  • Memory-access patterns can leak structure.
  • Subtle, hard to detect, and easy to overlook.

The lesson: security isn't only about what a model says. What it *does* — and how fast, and where in memory — can also leak. Thorough red teaming looks at both.