New Darrow Research Reveals AI Agents Replicate Privacy Defects Linked to Historical $115M Class-Action Settlements
DLAB tests 404 AI-built applications to give commercial insurance carriers a data-backed view of AI-generated legal
Press Release Disclaimer: This is a press release distributed through the XPR Media network. It has not been independently verified by our newsroom.
NEW YORK, Sept. 30, 2026 (GLOBE NEWSWIRE) — Darrow, a technology company and AI lab specializing in litigation exposure management, today released new research from its Darrow Legal Alignment Benchmark (DLAB). While traditional benchmarks grade what a model knows about the law, DLAB gives commercial insurance carriers the industry’s first empirical benchmark to measure what AI agents actually do under legal constraints, demonstrating that AI legal exposure can be measured and underwritten rather than carved out via broad policy exclusions.
By testing 404 applications built by seven AI coding agents, DLAB exposes a critical blind spot for commercial cyber and E&O insurers. 30% of AI builds that followed expert privacy specifications and showed a working ‘Opt-Out’ button still loaded tracking scripts after users rejected them. That defect pattern is closest to statutory wiretap (CIPA) claims, which have historically settled for up to $115M at enterprise scale.
By benchmarking these defects against historical class-action settlements, Darrow gives insurance innovation officers and chief underwriters the data needed to price live software risk and expand coverage options.
“For six years, we’ve built a legal intelligence platform grounded in real court filings and historical settlement data to understand how corporate legal exposure forms,” said Evyatar “Evya” Ben Artzi, Co-Founder and CEO of Darrow. “With enterprise software development moving to AI agents, we launched DLAB to answer the defining question of this decade: not whether a system knows the law, but whether it actually obeys it when it acts.”
4 Key Research Findings:
- The “Ghost Compliance” Trap (Runtime Failure): Even when following expert privacy specifications, 30% of AI builds with a working opt-out kept loading tracking scripts after users rejected them, mirroring a defect pattern closest to California Invasion of Privacy Act (CIPA) wiretap and CCPA claims despite appearing compliant at the UI level.
- Model Choice Matters (Claude vs. Others): In this study, tool choice shaped reliability. Given the same expert specification, 20 of 30 Claude builds achieved >95% legal compliance, compared to just 2 of 34 builds from the other models tested.
- Historical Settlement Benchmarks: Darrow benchmarked the defects against historical wiretap and privacy settlements: $0.5M – $3.2M for classes of 35K–460K members, up to $115M for classes of 5M–220M.
- Policy vs. Specification Gap: High-level corporate guidelines fail to prevent code defects because naming a risk is different work from specifying the fix. Handing agents clear technical specifications lifted compliance on high-litigation parameters from 62% to 92.6%.
By evaluating what autonomous AI agents actually build rather than how they perform on written exams, DLAB provides commercial insurance carriers with the clarity needed to move past broad exclusions, reward well-controlled policyholders, and write AI risk with confidence. To read the complete study, visit www.darrow.ai/resources/dlab.
About the Research
Darrow analyzed 404 AI-built applications by seven AI coding agents from Anthropic, OpenAI, Google, GitHub, and Cursor. Researchers stress-tested each build across sensitive verticals like pediatric health and consumer finance to measure real-world legal compliance. DLAB is the first in an ongoing series of research evaluating AI legal alignment across commercial domains. To read the complete study, visit www.darrow.ai/resources/dlab.
About Darrow
Darrow is a technology company and AI lab studying how legal exposure forms across industries, markets, and regulatory environments — combining large-scale legal data, advanced AI systems, and deep domain expertise to surface emerging risk before it escalates. This research powers Darrow’s Legal Intelligence platform and Privacy Radar, used by law firms, carriers and compliance teams to identify risk early, assess it with greater confidence, and act decisively. To date, Darrow’s intelligence platform has identified over $22 billion in actionable legal risk. Learn more at www.darrow.ai.
Media Contact: Linda Rigano | linda@theriganogroup.com | 914-815-0396
