Why Are We Still Storing PII in 2026?
The Mercor Breach Is the Wake-Up Call — But Not the First
In April 2026, AI startup Mercor—valued at $10B—was hit by a major data breach that exposed candidate profiles, PII, employer data, and even source code. (TechCrunch)
The root cause wasn’t a simple hack. It was a supply chain compromise—a poisoned open-source dependency (LiteLLM) that cascaded across thousands of systems in minutes. (TechCrunch)
Up to 4TB of data was reportedly exfiltrated. (Asanify)
And once again, the same uncomfortable question surfaces:
Why are we still storing PII at this scale in the first place?
The Real Problem Isn’t Security — It’s Architecture
Every breach gets framed as a security failure.
But Mercor exposes something deeper:
This is an architecture failure.
Companies still operate on a legacy assumption:
Collect user identity
Store it centrally
Secure it as best as possible
That model made sense in 2005.
It does not make sense in:
AI-native systems
Multi-agent environments
Global data supply chains
Because today:
Data flows across vendors, APIs, agents, and pipelines
One compromised dependency = full system exposure
Identity becomes the most valuable—and most vulnerable—asset
The Hidden Reality: PII Is a Liability, Not an Asset
The industry still treats PII like oil.
But in reality, PII behaves more like radioactive material:
Expensive to store
Dangerous to transport
Catastrophic when leaked
In the Mercor case:
Social Security numbers
Identity documents
Video interviews
Behavioral data
All became attack surface. (Compliance Hub)
And once exfiltrated:
You cannot rotate it (unlike passwords or API keys)
You cannot revoke it
You cannot contain its downstream impact
So Why Do Companies Still Store PII?
1. Legacy Compliance Thinking
Most systems are built around:
KYC / IDV storage requirements
Audit trails tied to raw identity
But compliance frameworks haven’t caught up to:
Tokenization
Zero-PII architectures
Decentralized identity models
2. Data Monetization Incentives
PII fuels:
Targeting
Analytics
AI training
But this creates a perverse incentive:
The more data you store, the more risk you accumulate.
3. Broken Identity Models
Today’s identity stack is still:
Username / password
Session-based
Database-backed
Even “modern” IAM systems still rely on:
Centralized identity storage as the source of truth
4. AI Made It Worse
AI systems require:
Massive datasets
Cross-system integrations
Continuous data flows
This means:
More PII collected
More places it’s stored
More surfaces to attack
The Mercor breach wasn’t just a company issue—it was an AI ecosystem failure.
The Shift: From Storing Identity → Verifying Identity
The future isn’t better encryption.
It’s not storing PII at all.
The New Model:
Identity is verified once
A tokenized identity (HIT) is issued
Systems operate on tokens—not raw PII
Access, actions, and AI execution are bound to that token
This changes everything:
No centralized PII honeypot
No reusable identity data to steal
No cascading breach impact
Why This Matters Now
The Mercor incident triggered:
Lawsuits
Enterprise contract suspensions
Industry-wide trust concerns
This is the pattern:
Breach
PR response
Credit monitoring
Repeat
What doesn’t change?
The underlying architecture.
The Bottom Line
We don’t have a breach problem.
We have a data model problem.
As AI systems scale:
Agents act autonomously
Data moves dynamically
Trust must be programmable
The only viable path forward is:
Identity without exposure. Execution without PII. Governance without centralized storage.
Closing Thought
The question is no longer:
“How do we secure PII?”
The real question is:
“Why does your system still need to store it?”
