Third-Party AI Audits

Independent testing and assurance of AI systems.

New York State auditors found 17 potential violations of NYC's hiring-AI audit law where the city regulator found one, and the city received two complaints in two years. US CAISI has testing agreements with five frontier labs, and the AICPA launched a combined SOC 2 and agent-testing report on October 19, 2026.

NYC Local Law 144 enforcement audit: Companies reviewed 32; Potential violations, state auditors 17; Violations found by city regulator 1; Complaints to city, two years 2 NYC LOCAL LAW 144 ENFORCEMENT AUDIT Companies reviewed 32 Potential violations, state auditors 17 Violations found by city regulator 1 Complaints to city, two years 2 COUNT, DEC 2025
Labs with CAISI agreements
5
CAISI evaluations completed
40+
CAICT models rated high risk
2 of 15
SB 53 penalty
$1M

Evidence

The Big Four and the AICPA entered AI assurance as government testing expanded

Developments: third-Party Audits
DateDevelopmentCountryWhat happenedSource
Jul 2025CAICT safety evaluationChina15 models tested; nine medium risk, two highSource
Aug 1, 2025PwC Assurance for AIUnited KingdomFirst Big Four AI assurance launchSource
Dec 2, 2025NY Comptroller audit of LL144United StatesWeak enforcement by the city regulatorSource
May 5, 2026CAISI agreements expandedUnited StatesGoogle DeepMind, Microsoft and xAI joined OpenAI and AnthropicSource
Oct 19, 2026AICPA and AIUC joint reportUnited StatesSOC 2 plus AIUC-1 agent testing in one engagementSource

Rules

Audit duties exist for hiring tools and frontier models; most assurance is voluntary

Key rules and standards: third-party audits
RuleJurisdictionWhat it requiresStatus
NYC Local Law 144New York CityAnnual independent bias audit of hiring toolsIn force since Jul 2023
California SB 53CaliforniaDescribe third-party catastrophic-risk assessmentIn force Jan 1, 2026
EU AI Act notified bodiesEUThird-party conformity assessment for some high-risk AIDec 2027
AIUC-1 with SOC 2US industryAgent testing with quarterly retestsLaunched Oct 2026
CAICT AI safety and trustworthy AI evaluationsChinaVoluntary certificates after testing2026 round announced Mar 2026

Status colors: green in force, blue voluntary standard or framework, amber adopted and phasing in, gray draft or pending, red withdrawn.

Record

Page Record

Datasets
ai_compliance_evidence.csv (5 rows)
Key sources
CAICT safety evaluation
PwC Assurance for AI
NY Comptroller audit of LL144
CAISI agreements expanded
AICPA and AIUC joint report
Last verified
October 2026