Mar 10, 2023 - Technology

Data spills in the age of AI

Ilustration: Rebecca Zisser / Axios

This week's tech headlines offer a compendium of data-loss nightmares:

  • The personal data of members of Congress was potentially exposed after hackers broke into a D.C. health-insurance system.
  • Police asked an Ohio businessman for video from his Ring doorbell camera, then issued a warrant for footage from more than 20 other cameras at his home and business.
  • Chinese-owned TikTok faces the threat of a ban over fears that the user data it collects could get fed to Beijing.
  • And users are sharing their mental health woes with OpenAI's ChatGPT with no concern for confidentiality.

What's happening: Congress' long-running inability to pass a comprehensive privacy law has left online personal information vulnerable to be mined, hoarded and poached.

Why it matters: Virtually every major technology today opens data vulnerabilities that can cause havoc.

  • "Data privacy" may sound like an abstraction to much of the U.S. public, but our national failure to set privacy rules can have very concrete consequences.

Zoom out: Legal experts and privacy advocates have long warned of the dangers of the U.S.'s failure to bring privacy law into the 21st century.

  • It means that government authorities have a freer hand to seize digital information as evidence.
  • Private companies are freer to gather and resell the personal information of their customers and users.
  • In both public and private sectors, the absence of tough rules governing data handling makes every breach and hack more potentially damaging.

What's next: The frenzy over generative AI is adding a whole new dimension of worry.

  • AI experts fear that chatbots like ChatGPT trained on vast troves of internet text will already be seeded with an unknowable volume of personal data.
  • On its own, that's little different from what's available on Google or any other search engine today.
  • The difference is that ChatGPT and similar programs are capable of "remembering" and reusing information users share with them in unpredictable ways.
  • That means that details from any legal document, medical report, financial calculation or other input that someone shares with these systems might turn up again — accurately or erroneously — in answers to someone else's query, with no indication of the original source.

Our thought bubble: Every time you type at ChatGPT, consider that you might be sharing secrets with a thing that has an impossibly vast memory — and doesn't have a clue what a secret even is.

Between the lines: There may well be ways to equip generative AI systems with guardrails to protect against this kind of unintended sharing.

  • But right now developers have little incentive to build them, and the rest of us have no visibility into what data the systems are holding onto.

The bottom line: The faster technology advances and the more central it becomes in our lives, the more we'll miss having a good privacy law.

Go deeper