AI
Security vs Privacy: Mindset and Adversary
Security engineers fight nation-state hackers. Privacy engineers fight data scientists with good intentions. Security bad means a breach or ransomware. Privacy bad means a failed deletion or child data in a model. Security thinks in big buckets like critical data. Privacy thinks in columns and rows: biometric, child data, accounting. Security wants to keep data. Privacy wants to lose it. These are not the same job.
Personal Data is Not PII
Security people call personal data PII. That is a trap. GDPR, CCPA, CP do not use the word PII. They talk about identifiable natural people. Personal data is anything that can identify a person directly or indirectly. A user ID or session ID counts. Even a UUID that links to a user is personal data. Scanners cannot find it easily because UUIDs look like random strings. Catalog in advance.
The Adversary: Well-Intentioned Data Scientists
In security, there is a bad guy. In privacy, there is almost never a bad guy. Your worst adversary is a very well-intentioned data scientist. They reidentify anonymous data. They keep data they should delete. They do not tell you what they have. They mean well. You must explain why their actions are dangerous to the company. It is a kinder, gentler practice than security.
Upgrade Your Data Catalog for Privacy
To add privacy to your security program, upgrade your data catalog. Collect additional metadata. Sit with lawyers to understand legal obligations. Catalog table by table what is personal data. Keep it up to date. Automate cataloging where possible. Use the 80/20 rule: automate what you can, call a lawyer for the rest. You need to know purposes: is this for ads, ML, tax? Model purpose in access control with roles or attributes.
Child Data: The Glitterbomb of Privacy
Child data is the glitterbomb of privacy hand of privacy data handling. Once it touches something, that thing is forever tainted. Regulators get upset. You must prove you handled it correctly. If you cannot, everything that might have touched child data is tainted. The FTC made Weight Watchers delete their entire trained ML model because they could not prove they had not used child data. Consent changes over time. Check consent every single time.
Deletion and Retention: Absolute and Irrevocable
Privacy requires absolute, irrevocable deletion. Hashing the user ID is not enough. Delinking sessions is not enough. Backups must be deleted after the compliance SLA, typically 15 to 60 days. Exceptions: tax and accounting data, fraud and security, litigation holds. You must know whether a system is allowed to keep data past deletion. Retention depends on purpose. Ads targeting has a short runway. Tax data lasts years. When a purpose expires, delete the data.
Q&A
How can a cyber practitioner learn to reduce privacy risks? Start with cataloging and metadata; Nishant Bargaria’s book is a good starter kit. ▶ 26:19
Is tax and accounting data about individuals or the organization? Only individual-related tax and accounting data; company’s own accounting is not a privacy problem. ▶ 27:18
How to trace user data through ML model training? Use data extraction attacks to test if model memorized data; train robustly against extraction. ▶ 28:27
Notable Quotes
Your worst adversary is a very well-intentioned data scientist. Aleatha Parker-Wood · ▶ 6:42
child data is the glitterbomb of privacy hand of privacy data handling Aleatha Parker-Wood · ▶ 17:57
privacy is all about losing data Aleatha Parker-Wood · ▶ 8:56
you can’t just go down a checkbox list and say, “Okay, GDPR, CCPA, HIPPA, I’m done.” Aleatha Parker-Wood · ▶ 15:06
Security engineers need to worry about nation state hackers. Privacy people need to worry about data scientists with very good intentions. Aleatha Parker-Wood · ▶ 24:34
Key Takeaways
- Privacy’s adversary is the well-intentioned data scientist, not a malicious hacker.
- Personal data is broader than PII; any identifier that links to a person counts.
- Privacy requires granular data catalogs, purpose-based access control, and absolute deletion.
About the Speaker(s)
Dr. Parker-Wood is currently the founder and principal consultant at Global Optima LLC, which provides technical/executive consulting on security, privacy, and AI governance.