1. Who Audited Your Auditor?
A $32 million compliance startup fabricated the security certifications your vendor assessment depends on. The cascade that followed exposed how thin the trust layer beneath the AI supply chain actually is.
A $32 million compliance startup fabricated the security certifications your vendor assessment depends on. The cascade that followed exposed how thin the trust layer beneath the AI supply chain actually is.
Delve Technologies sold automated SOC 2 (the audit standard companies use to prove their security controls work) and ISO 27001 (international information security certification) compliance. Y Combinator, the startup accelerator whose backing is itself a credibility signal, funded it. Hundreds of AI companies bought it. An anonymous researcher who accessed a leaked Google Spreadsheet of Delve's draft reports found that 493 of 494 used identical boilerplate text, with only the client's name and logo swapped in. Auditor conclusions and test results were fully populated before clients submitted company descriptions, network diagrams, or evidence. The certifying firms were Indian entities operating through US shell companies, not the recognized CPA firms the reports implied. Y Combinator expelled Delve in early April.
One of those certified companies was LiteLLM, an open-source AI gateway present in roughly 36 percent of cloud environments. On March 24, attackers compromised a security scanner in LiteLLM's build pipeline, stole the token used to publish software packages, and pushed two malicious versions to PyPI (Python's public software repository). The poisoned code ran for 40 minutes before it was quarantined. In that window, Mercor, a $10 billion AI staffing platform that generated training data for OpenAI, Anthropic, and Meta, auto-installed the compromised package. Attackers exfiltrated four terabytes: platform source code, 40,000 contractor records including Social Security numbers, video interview recordings with passport scans, and internal communications. Meta reportedly froze its AI data work with Mercor after the breach.
Five lawsuits landed within a week. The one that matters: White and Beltran v. Mercor (Northern District of Texas) names Delve and LiteLLM's parent company as co-defendants alongside Mercor. The legal theory connects the compliance fraud to the breach to the data loss as a single chain of liability. A vendor's fake certification made the next vendor's security unverifiable, which made the platform's data unprotectable.