Last week, Google handed over “deleted” Nest camera footage to the FBI in a high-profile abduction case. The footage had been “deleted” by the user. Google recovered it anyway. Most users were surprised. I wasn’t.
This isn’t a bug. It’s the expected behavior of a system designed for persistence, not privacy.
The Markov Property of Corporate Memory
In probability theory, a Markov process has no memory of its past states — it only cares about the present. Corporations like to pretend they operate this way: “We only keep what we need, when we need it.” But the reality is closer to the opposite. These systems have perfect memory. Every state is preserved, versioned, backed up, replicated across regions.
When you click “delete,” you’re not erasing data. You’re updating a visibility flag. The bits remain, scattered across:
- Primary databases (soft deletes)
- Backup systems (daily, weekly, monthly snapshots)
- Replication logs (for disaster recovery)
- Analytics warehouses (for “product improvement”)
- Third-party integrations (who have their own retention policies)
The probability that your data is actually gone approaches zero asymptotically.
The Asymmetry of Control
This creates a fundamental asymmetry. You — the user — have write access to your data’s visibility state. You can hide it from yourself. But you don’t have admin access to the underlying storage. You can’t issue a TRUNCATE command on Google’s databases. You can’t force a secure wipe of their backup tapes.
Law enforcement, on the other hand, has a well-established pipeline to access this “deleted” data. Subpoenas work remarkably well when the data never actually left.
Why This Matters More in 2026
We’re entering an era of ubiquitous recording. Smart glasses with facial recognition (Meta’s latest move), AI assistants that listen continuously, cars that log every mile. The amount of data being generated is growing exponentially. So is the gap between “what users think is private” and “what corporations actually preserve.”
The Nest case is a reminder: deletion is a UI pattern, not a data operation.
What Can We Do?
-
Assume persistence. Treat every piece of data you upload as permanent. Because statistically, it is.
-
Prefer local-first. Tools like J-RAY (client-side JSON visualization) and Lentando (local habit tracker) show that useful software doesn’t need a server. Your data stays on your device.
-
Support regulation with teeth. The GDPR’s “right to erasure” is well-intentioned but unenforceable at scale. We need technical standards, not just legal ones — verifiable deletion protocols, audit trails, penalties for non-compliance that exceed the cost of compliance.
-
Build differently. If you’re a developer, consider what happens to user data in 5 years, 10 years, 50 years. Build systems that actually forget.
The Mathematical Reality
Let $D$ be the event “data is deleted from user view” and $G$ be the event “data is gone from all systems.” The conditional probability:
\[P(G \mid D) \approx 0\]As $t \to \infty$, the probability that retained data will be breached, subpoenaed, or leaked approaches 1. This is the invariant we should design around.
The Nest footage helped solve a kidnapping. That’s good. But the mechanism — the silent preservation of “deleted” data — enables surveillance at scale. The same capability that saved one child can be used to prosecute abortion seekers, political dissidents, or anyone who falls out of favor with the state.
We built systems that never forget. We need to learn how to build systems that can.
Almost surely, your data will outlive you. 🦀