In January 2025, Sax LLP, a top-100 US accounting firm with annual revenue exceeding $100 million, began notifying over 228,000 individuals that their sensitive data had been compromised. The breach had gone undisclosed for more than 16 months. Hackers gained access to systems in late July 2024 and exfiltrated files containing Social Security numbers, financial account information, and tax records before Sax detected the intrusion on August 7th.
The investigation took another year after detection. While Sax investigated, those 228,000 people had no idea their information was circulating in criminal marketplaces.
This wasn't an isolated incident. Chicago-based Legacy Professionals LLP notified 216,752 individuals after a 2024 hack and now faces at least five class-action lawsuits. Wojeski & Company in Albany paid a $60,000 settlement to the New York Attorney General after two ransomware attacks exposed 4,700 people. The AG's office found the firm took over a year to notify victims.
The accounting industry has become the target. Since 2020, reported attacks on accounting practices have increased 300%. In 2024, the IRS received over 250 data breach reports from tax professionals, impacting more than 200,000 clients. Accounting firms face an average of 900 cyberattack attempts during tax season alone. According to IBM's Cost of a Data Breach 2024 report, financial services breaches now cost an average of $6.08 million, 22% above the global mean.
And now those same firms are adopting AI to process the very documents that make them targets.
The short version: If you need to redact sensitive documents before they reach AI systems, PaperVeil handles that layer. The rest of this article explains where it fits in the broader governance architecture.
The Data Landscape: What Accounting Firms Hold
Accounting firms are concentrated repositories of some of the most valuable data in existence. A single client engagement can generate:
Tax Information:
- Social Security numbers for every family member
- Employer identification numbers
- Prior year tax returns with full financial history
- W-2s, 1099s, K-1s, and other income documentation
- Deductions, credits, and financial positions
Financial Records:
- Bank account numbers and statements
- Investment account information
- Credit card statements
- Loan documentation
- Net worth calculations
Business Client Data:
- Payroll records with employee SSNs
- Financial statements and projections
- Accounts receivable and payable details
- Vendor and customer information
- Merger and acquisition documentation
- Corporate tax returns with owner information
Personal Client Information:
- Names, addresses, dates of birth
- Driver's license numbers
- Contact information for family members
- Estate planning documents
- Trust documentation
A successful breach at an accounting firm doesn't just expose one person. It exposes entire families, business owners and their employees, and interconnected networks of financial relationships. The 2024 breach at a 12-person Midwest firm illustrates this. Attackers exfiltrated three months of client data undetected. The breach was only discovered when two clients reported fraudulent tax returns filed in their names. By 2025, the firm had dissolved. Trust was broken, and clients walked away.
The AI Adoption Pressure
The pressure to adopt AI in accounting is substantial and growing. Tax preparation involves massive document processing workloads concentrated in compressed timeframes. AI promises to:
- Extract data from uploaded documents automatically
- Draft engagement letters and correspondence
- Summarize complex financial positions
- Research tax code questions
- Process routine client inquiries
- Review contracts and agreements
- Generate initial versions of financial statements
The productivity gains are real. A staff accountant using AI assistance can process documents faster, research questions more efficiently, and handle more clients during busy season. Firms that don't adopt AI risk falling behind competitors who do.
But every document uploaded to an AI system becomes a potential exposure point. When a staff member pastes a client's tax return into ChatGPT to get help with a complex deduction question, that document, with its Social Security numbers, income figures, and financial positions, now exists on OpenAI's servers. The training toggle doesn't prevent transmission. It only prevents future model training. The data still goes somewhere your firm doesn't control.
The Verizon Data Breach Investigations Report (DBIR 2024) found that 74% of breaches involve the human element: phishing, stolen credentials, or misused accounts. Staff members using unsanctioned AI tools fall squarely into this category.
Risk Matrix: Data Types and Exposure Levels
Not all accounting data carries equal risk. Here's how to think about the exposure levels:
Critical Risk (Immediate Financial Harm):
- Social Security numbers
- Bank account and routing numbers
- Credit card numbers
- Investment account credentials
Exposure enables immediate identity theft, fraudulent tax returns, and direct financial theft. These identifiers never expire. A stolen SSN is compromised permanently.
High Risk (Significant Harm Potential):
- Complete tax returns
- Financial statements with account details
- Payroll files with employee SSNs
- Business financial projections
- M&A documentation
Exposure enables sophisticated fraud, competitive harm, and cascading breaches affecting multiple individuals.
Medium Risk (Privacy Violation, Regulatory Exposure):
- Names and addresses
- Income levels without account numbers
- Business relationships
- Engagement letters
Exposure violates privacy and triggers notification requirements but creates less direct harm pathway.
Lower Risk (Operational Data):
- Anonymized financial analysis
- General tax research
- Internal process documentation
Exposure creates reputational risk but limited direct client harm.
This framework determines what can safely interact with AI systems and what requires redaction or exclusion.
Security Architecture: What Good Looks Like
Effective document security for accounting firms in the AI era requires multiple layers:
Layer 1: Classification at Intake
Every document entering the firm should be classified by sensitivity level. This can't be manual for every document, but it can be systematic:
- Client tax returns: Automatically classified as Critical
- Bank statements: Automatically classified as Critical
- Engagement letters: Automatically classified as Medium
- General correspondence: Case-by-case or Lower
Classification drives handling rules. Critical documents never go to external AI without redaction. High-risk documents require specific approval. Lower-risk documents may have more flexibility.
Layer 2: Access Controls
Not everyone needs access to everything. Implement:
- Role-based access to client files
- Logging of all document access
- Separate credentials for sensitive systems
- Multi-factor authentication universally
- Session timeouts for inactive connections
The principle: minimize the blast radius of any compromised account.
Layer 3: AI Governance
Create clear policies for AI use:
- Approved AI tools only (enterprise tiers with DPAs, not personal accounts)
- Mandatory redaction for any document containing Critical data before AI processing
- No client names, SSNs, account numbers, or identifying information in AI prompts
- Logging of AI interactions for audit purposes
- Regular review of AI usage patterns
Layer 4: Pre-Upload Redaction
For documents that need AI processing, implement automated redaction:
- Detect and remove SSNs, EINs, and other tax identifiers
- Strip names and replace with placeholders
- Remove account numbers and routing numbers
- Eliminate addresses and contact information
- Generate audit trail of what was redacted
The AI receives anonymized content. You get the analytical benefits without the exposure.
Layer 5: Monitoring and Response
- Network monitoring for unusual data movement
- Alerts for large file transfers
- Incident response plan tested annually
- Cyber insurance appropriate to exposure level
- Relationship with forensics firm established before you need them
Implementation for Accounting Firms
Here's a practical path to implementation:
Phase 1: Assessment (Weeks 1-2)
Inventory your current state:
- What AI tools are staff using? (Survey honestly; the answer may surprise you)
- What client data exists in your systems?
- What's your current document handling workflow?
- Who has access to what?
Phase 2: Policy Development (Weeks 3-4)
Create clear, enforceable policies:
- Approved AI tools list
- Data classification scheme
- Handling requirements by classification level
- Consequences for policy violations
Phase 3: Technical Implementation (Weeks 5-8)
Deploy technical controls:
- Enterprise AI accounts with proper DPAs
- Redaction tooling integrated into workflow
- Access logging and monitoring
- MFA everywhere
Phase 4: Training (Week 9)
Train all staff on:
- New policies and why they exist
- How to use approved AI tools safely
- What to do if they're uncertain
- How to report potential incidents
Phase 5: Ongoing Operations
- Quarterly policy review
- Annual penetration testing
- Regular staff refresher training
- Continuous monitoring and improvement
Compliance Mapping
Accounting firms operate under specific regulatory requirements:
FTC Safeguards Rule (16 CFR Part 314): Applies directly to CPA firms and tax professionals. Requires written security program, risk assessment, employee training, and incident response capability. The FTC penalized a tax preparation company in 2024 for failing to encrypt client data under this rule.
Gramm-Leach-Bliley Act (GLBA): Covers financial institutions including accounting firms handling client financial data. Requires privacy notices and data safeguards.
IRS Publication 4557: Provides data security guidance specifically for tax professionals. While not directly enforceable, following it demonstrates reasonable security practices.
State Data Breach Notification Laws: All 50 states plus DC have breach notification requirements. Timelines vary, but the Wojeski case shows New York takes delayed notification seriously.
State CPA Board Requirements: Many state boards have adopted ethics rules addressing data protection and technology use.
AICPA Professional Standards: Professional standards require protecting client confidentiality. Using consumer AI tools with client data arguably violates these standards.
The Choice
Accounting firms face a genuine tension. AI offers meaningful productivity gains in a profession facing staffing shortages and seasonal workload compression. But the same documents that AI could help process contain exactly the information criminals want most.
The solution isn't avoiding AI. It's implementing AI thoughtfully, with proper controls, proper tools, and proper safeguards. Redaction before AI processing eliminates the core risk while preserving the analytical benefits.
The alternative is hoping you don't become the next headline. Given the trajectory of attacks against accounting firms, hope is not a viable security strategy.
PaperVeil lets you redact all your sensitive information from PDFs in a simple drag and drop flow. Detect and remove PII, match custom patterns, strip metadata, and generate audit trails. The redaction layer that makes AI document processing actually safe.