What are the key ongoing obstacles that make data anonymization a hard, unsolved problem?

Preventing re-identification, balancing privacy vs. usability, the lack of global standards, and AI/ML cutting both ways.

Anonymization is "unsolved" not because the techniques are weak but because it has to satisfy several demands that pull against each other at once, and one of the forces — the attacker — keeps getting stronger. Four obstacles capture why there's no permanent finish line:

• Preventing re-identification — the core technical problem never fully closes. Modern ML detects subtle patterns in supposedly anonymized data, and ever-growing linkage methods merge datasets, so re-association risk persists despite best efforts.
• Privacy vs. data usability — the unavoidable trade-off. Push privacy up and you destroy utility; keep utility and you risk re-identification. The right balance is context-dependent: sensitive medical records demand far more rigorous anonymization than low-risk demographic data.
• Establishing global standards — a governance gap, not a math one. As data's value grows, uniform oversight is increasingly needed, yet rules differ across jurisdictions; even a robust regime like the GDPR can make legitimate sharing harder.
• AI & ML cut both ways — the same technology is both shield and weapon. It can help (detect PII automatically, generate synthetic data via GANs) but also attack (assist de-anonymization and re-identification), so advances rarely give a lasting net advantage to the defender.

Tip: Anonymization is a moving equilibrium between three forces — privacy, utility, and an attacker whose tools keep improving. There's no permanent "solved" state.