SOC 2 Compliance for AI Voice Agents: What You Need to Know

TL;DR

SOC 2 is a security auditing framework developed by the American Institute of Certified Public Accountants (AICPA). It evaluates how a service provider handles customer data across five Trust Service Criteria: security, availability, processing integrity, confidentiality, and privacy. For AI voice agents that process sensitive phone conversations, SOC 2 Type II certification means an independent auditor has verified that security controls are not just designed but actually operating effectively over a minimum 6-month period. When evaluating AI voice vendors, ask for the full SOC 2 Type II report - not just a badge on a website - and verify that the audit scope covers the voice processing components you will actually use.

Trust Service Criteria

6-12 mo

Type II Observation Period

3-9 mo

Typical Audit Timeline

$50-200K

Average Audit Cost

When your AI voice agent handles customer phone calls, it processes names, phone numbers, health information, payment details, and appointment data in real time. The question is not whether this data needs protection - it does - but how you verify that your vendor's security claims are more than marketing copy.

SOC 2 provides that verification. It is not a self-certification or a checklist a vendor fills out themselves. It is an independent audit conducted by a licensed CPA firm that examines actual security controls, tests them, and reports on whether they work as described. For businesses choosing an AI voice platform, understanding SOC 2 is essential for making an informed vendor decision.

What Is SOC 2 and Why It Matters for Voice AI

SOC 2 (System and Organization Controls 2) is an auditing standard created by the AICPA specifically for service organizations that store, process, or transmit customer data. Unlike compliance frameworks that are prescriptive (HIPAA tells you exactly what to do), SOC 2 is principles-based - it defines what outcomes your security controls must achieve but gives you flexibility in how you achieve them.

For AI voice agents, SOC 2 matters for three reasons:

Voice data is uniquely sensitive: Phone conversations contain biometric data (voiceprints), personally identifiable information (PII), and often protected health information (PHI). The combination of data types in a single voice call is richer than almost any other business application.
Multi-system architecture creates attack surface: A single AI voice call touches telephony infrastructure, speech-to-text engines, large language models, text-to-speech systems, CRM integrations, and data storage. Each connection point must be secured.
Enterprise buyers require it: If you are deploying AI voice agents for a mid-market or enterprise client, their procurement team will ask for SOC 2. Without it, you are excluded from consideration regardless of how good your technology is.

SOC 2 Type I vs Type II: The Difference That Matters

There are two types of SOC 2 reports, and the distinction matters significantly:

Aspect	SOC 2 Type I	SOC 2 Type II
What it evaluates	Design of controls at a point in time	Design AND operating effectiveness over time
Observation period	Single date (snapshot)	Minimum 6 months, typically 12 months
What it proves	Controls exist on paper	Controls actually work consistently
Time to complete	1-3 months	6-12 months minimum
Cost	$20,000-80,000	$50,000-200,000
Enterprise acceptance	Often insufficient	Standard requirement
Renewal	One-time assessment	Annual re-audit required

Type I says: "On March 1, 2026, these controls were in place." Type II says: "From March 1 to September 1, 2026, these controls were in place AND they operated effectively every day."

The practical difference is enormous. A vendor can set up controls the day before a Type I audit, pass, and then ignore them afterward. Type II requires sustained evidence of consistent operation. For AI voice agents handling your customers' phone conversations daily, you want Type II assurance that security controls work every day - not just on the day the auditor visited.

Watch Out for Type I Only

Some vendors advertise "SOC 2 certified" or display SOC 2 badges without specifying the type. Always ask whether it is Type I or Type II. A Type I report is a reasonable starting point for a young company, but any established AI voice platform handling production data should have Type II certification. If a vendor has been operating for more than 18 months and only has Type I, ask why.

The Five Trust Service Criteria Applied to Voice AI

SOC 2 evaluates controls across five Trust Service Criteria (TSC). Not all five are required - security is mandatory, and the others are selected based on the service. For AI voice agents, all five are relevant:

1. Security (mandatory - the Common Criteria)

Security is the foundation and is always included. For AI voice platforms, security controls cover:

Access controls to the voice processing infrastructure
Network security and firewall configurations
Encryption of voice data in transit (TLS/SRTP) and at rest (AES-256)
Vulnerability management and penetration testing
Incident detection and response procedures
Change management for AI model updates and system changes

2. Availability

Availability evaluates whether the system is operational and usable as committed. For an AI voice agent that is your business's phone system, downtime means missed calls and lost revenue. Availability controls include:

Uptime SLAs and monitoring systems
Disaster recovery and business continuity plans
Failover architecture and redundancy
Capacity planning for call volume spikes
Backup telephony routing if AI systems are unavailable

3. Processing Integrity

Processing integrity ensures the system processes data completely, accurately, and in a timely manner. For AI voice agents, this means:

Speech-to-text accuracy monitoring and quality controls
Correct routing of calls and data to intended destinations
Accurate appointment booking with no data corruption
Complete and accurate call logging and transcript generation
Error detection and correction mechanisms

4. Confidentiality

Confidentiality controls protect information designated as confidential. In voice AI:

Call recordings accessible only to authorized personnel
Business knowledge bases isolated between clients (multi-tenancy security)
Encryption key management and rotation policies
Data classification and handling procedures
Non-disclosure and confidentiality in vendor contracts

5. Privacy

Privacy covers the collection, use, retention, disclosure, and disposal of personal information. For voice AI:

Privacy notices communicated to callers
Consent management for call recording
Data retention and deletion policies
Data subject access request (DSAR) handling
Cross-border data transfer safeguards

What SOC 2 Auditors Examine in an AI Voice Platform

A SOC 2 audit is not a theoretical exercise. Auditors examine actual evidence across dozens of control areas. For an AI voice platform, this typically includes:

Infrastructure security review

Auditors examine cloud configurations (AWS, GCP, Azure), network architecture, firewall rules, and intrusion detection systems. For voice AI, they specifically review the security of WebSocket connections used for real-time audio streaming and the SIP/SRTP configuration for telephony.

Access management testing

Every employee with access to production systems is examined. Auditors verify multi-factor authentication, role-based access controls, regular access reviews, and proper offboarding when employees leave. They test whether former employees still have access.

Encryption verification

Auditors verify that encryption claims are real - they check TLS certificates, SRTP configurations, AES-256 key management, and encryption of data at rest in databases and object storage. They may perform packet captures to verify audio streams are actually encrypted.

Change management evidence

Every change to production systems - code deployments, AI model updates, infrastructure changes - must go through a documented change management process. Auditors review change logs, approval workflows, and rollback procedures.

Incident response testing

Auditors review incident response plans and evidence of tabletop exercises or actual incident handling. They examine whether incidents were detected, responded to, and resolved within documented timeframes.

Vendor management review

For AI voice platforms that use sub-processors (cloud providers, LLM APIs, telephony providers), auditors examine how these vendors are assessed, contracted, and monitored. This is the supply chain security component.

Data lifecycle documentation

From the moment a call begins to the moment data is deleted, auditors trace the complete data lifecycle. They verify that data retention policies are enforced, deletion procedures are executed, and data does not persist beyond stated retention periods.

The Sub-Processor Problem: SOC 2 in a Multi-Vendor Stack

Here is where SOC 2 for AI voice agents gets complicated. A typical AI voice platform relies on multiple sub-processors:

Component	Example Providers	Data They Process	SOC 2 Available?
Cloud infrastructure	AWS, GCP, Azure	All data in the platform	Yes - all three have SOC 2 Type II
Telephony / SIP	Telnyx, Twilio, Vonage	Call metadata, audio streams	Yes - major providers have SOC 2
Speech-to-text	OpenAI Whisper, Google STT, Deepgram	Audio recordings, transcripts	Varies - check each provider
Large language model	OpenAI, Anthropic, Google	Conversation context, prompts	Varies - evolving rapidly
Text-to-speech	ElevenLabs, OpenAI TTS, Google TTS	Response text	Varies - newer providers may lack it
Database	PostgreSQL on cloud, Supabase, Neon	All stored customer data	Depends on hosting configuration

A vendor's SOC 2 report covers their own controls - but it also must address how they manage these sub-processors. The concept of "complementary sub-service organization controls" (CSOCs) is how SOC 2 handles this. The report should clearly state which sub-processors are used, what controls the vendor relies on them to provide, and how the vendor monitors those sub-processors.

When reviewing a vendor's SOC 2 report, look for any "carve-outs" - components excluded from the audit scope. If the telephony layer or LLM processing is carved out, the report does not cover the security of those components, and you need to assess them separately.

SOC 2 vs ISO 27001 vs GDPR: How They Relate

Aspect	SOC 2	ISO 27001	GDPR
Type	Audit report	Certification	Legal regulation
Origin	AICPA (US)	ISO (International)	EU Parliament
Focus	Service organization controls	Information security management	Personal data protection rights
Mandatory?	No - market-driven	No - voluntary	Yes - for EU data processing
Scope	Technology controls and processes	Organization-wide security management	Any processing of EU personal data
Audit frequency	Annual	Surveillance audits + 3-year recertification	Ongoing enforcement
Relevance for US clients	Primary standard	Recognized but less common	Required if processing EU data
Relevance for EU clients	Valued addition	Primary standard	Legal requirement

These frameworks complement each other. SOC 2 provides detailed evidence of security controls that auditors can verify. ISO 27001 establishes an information security management system across the organization. GDPR is a legal obligation for processing EU personal data. The strongest position for an AI voice vendor serving international clients is to hold all three.

How to Assess a Vendor's SOC 2 Report

When a vendor provides their SOC 2 report (often under NDA), here is how to evaluate it:

Check the report type and period: Confirm it is Type II and note the observation period. A report covering January-December 2025 is current in early 2026 but a report from 2023 is stale and should raise questions about whether the vendor maintained compliance.
Review the system description: Section III of the report describes the system being audited. Verify it covers the voice AI components you will actually use - not just a billing portal or marketing website.
Look for exceptions: Section IV contains the auditor's test results. Any exceptions (controls that did not operate effectively) are listed with management's response. One or two minor exceptions are normal. Numerous exceptions or exceptions in critical security controls are red flags.
Check sub-processor handling: Review how sub-processors are addressed. Are they included in the audit scope (inclusion method) or carved out? Carve-outs for critical components like cloud infrastructure or LLM providers mean those components are not audited.
Verify the auditor: The CPA firm conducting the audit should be reputable. Major firms (Deloitte, EY, KPMG, PwC, BDO, Grant Thornton) and specialized SOC 2 audit firms are trustworthy. An unknown firm raises questions about audit rigor.

10 SOC 2 Questions to Ask Any AI Voice Vendor

Do you have a current SOC 2 Type II report?

Not Type I, not "in progress," not "planned." A current Type II report with an observation period ending within the past 12 months. If they say "SOC 2 compliant" without specifying Type II, ask for clarification.

What is the audit scope?

Does the report cover the AI voice processing platform, or is it limited to a subset of systems? The audit must cover the components that process your customer call data.

Which Trust Service Criteria are included?

Security is always included. Ask whether Availability, Processing Integrity, Confidentiality, and Privacy are also covered. For voice AI, all five are relevant.

Are there any exceptions in the current report?

Vendors with clean reports will tell you proudly. Vendors with significant exceptions may be evasive. Ask specifically about exceptions and remediation status.

How are sub-processors handled?

Are cloud infrastructure, LLM providers, and telephony providers included in the audit scope or carved out? Carve-outs in critical components weaken the report.

Can we review the full report under NDA?

A legitimate SOC 2 report is shared under NDA with prospective and current clients. If a vendor refuses to share the report, their SOC 2 claim is unverifiable.

Who conducted the audit?

The auditing firm should be a licensed CPA firm with SOC audit experience. Ask for the firm name and verify their reputation independently.

How do you handle the gap between audit periods?

SOC 2 reports cover a defined period. Ask what controls and monitoring continue between audit periods and how they bridge any gap.

What changes have been made since the last audit?

If the vendor has significantly changed their architecture, added new sub-processors, or migrated infrastructure since the last audit, the report may not reflect current reality.

When is the next audit period?

Annual re-audits demonstrate ongoing commitment. If the vendor cannot tell you when their next audit starts, their compliance program may be inconsistent.

SOC 2 Implementation Roadmap for AI Platforms

If you are building or operating an AI voice platform and need SOC 2 certification, here is the typical roadmap:

Phase	Duration	Key Activities
Readiness assessment	4-6 weeks	Gap analysis against Trust Service Criteria, identify control deficiencies, remediation planning
Control implementation	2-4 months	Build and deploy controls: access management, encryption, logging, incident response, change management
Evidence collection setup	2-4 weeks	Configure continuous monitoring, automated evidence collection, policy documentation
Type I audit (optional)	4-6 weeks	Point-in-time assessment to validate control design before committing to Type II observation
Type II observation period	6-12 months	Controls operate and evidence accumulates. This period cannot be shortened.
Type II audit	4-8 weeks	Auditor reviews evidence, tests controls, interviews staff, produces report
Ongoing compliance	Continuous	Maintain controls, collect evidence, prepare for annual re-audit

Total timeline from zero to a SOC 2 Type II report: 12-18 months. This is why asking a new AI voice startup for SOC 2 Type II is a meaningful filter - it requires sustained investment in security infrastructure and processes that cannot be rushed.

Frequently Asked Questions

No. SOC 2 is not a legal requirement - it is a market-driven standard. However, many enterprise and mid-market buyers require SOC 2 Type II as a condition of doing business. In regulated industries like healthcare and finance, SOC 2 is often a practical prerequisite even though the legal requirement is technically HIPAA or PCI DSS.

SOC 1 focuses on financial reporting controls - it is relevant for service providers that affect a client's financial statements (payroll processors, payment platforms). SOC 2 focuses on information security and data protection controls. For AI voice agents, SOC 2 is the relevant standard unless the voice agent directly processes financial transactions.

It takes 12-18 months minimum from inception to achieve SOC 2 Type II. A company that launched 6 months ago cannot have it. They might have Type I or be in their observation period. This is not a disqualification, but you should understand the limitation and assess their security controls through other means.

SOC 2 means an independent auditor verified that security controls exist and operated effectively during the audit period. It significantly reduces risk but does not eliminate it. No certification guarantees zero breaches. SOC 2 is one important factor in vendor assessment, alongside technical architecture review, contractual protections, and ongoing monitoring.

Direct audit costs range from $50,000 to $200,000 annually depending on scope and auditor. Implementation costs (security tools, personnel, process changes) can add $100,000-500,000 in the first year. This investment is one reason SOC 2 serves as a useful quality filter for vendors.

Yes, though ISO 27001 is more commonly required in Europe. SOC 2 provides complementary assurance and is increasingly recognized by EU enterprises. If your AI voice vendor serves both US and EU markets, having both SOC 2 and ISO 27001 demonstrates comprehensive security coverage.

If a vendor fails their annual re-audit, their SOC 2 report expires and they can no longer claim current compliance. This should trigger a review of your vendor agreement and an assessment of the security implications. Most contracts include notification clauses for compliance status changes.

Traditional SOC 2 controls cover infrastructure security but AI-specific risks like prompt injection, model extraction, or training data poisoning are emerging concerns. Forward-thinking auditors are beginning to include AI-specific controls, but coverage varies. Ask specifically about AI security controls in addition to standard SOC 2 criteria.

Standard practice is to share SOC 2 reports with prospective clients under a non-disclosure agreement. If a vendor refuses to share their report before you sign a contract, this is a red flag. The report is the evidence behind the claim - without reviewing it, the SOC 2 badge is unverifiable marketing.

SOC 2 and HIPAA overlap significantly but are not identical. SOC 2 covers general security controls while HIPAA has specific requirements for protected health information (PHI). A healthcare AI voice vendor should ideally have both SOC 2 Type II and HIPAA attestation. SOC 2 with the Privacy criterion covers many HIPAA requirements but does not replace a dedicated HIPAA compliance assessment.

Justas Butkus

Founder & CEO, AInora

Building AI digital administrators that replace front-desk overhead for service businesses across Europe. Previously built voice AI systems for dental clinics, hotels, and restaurants.

View all articles

Ready to try AI for your business?

Hear how AInora sounds handling a real business call. Try the live voice demo or book a consultation.

Try Voice Demo Book Consultation