Scale AI Sues Ex-Employee and Rival Mercor Over Stolen Files

Scale AI has filed a lawsuit against a former sales employee, Eugene Ling, and rival data-labeling company Mercor, alleging the employee “stole more than 100 confidential documents” containing customer strategies and other proprietary information. The complaint, filed Wednesday, accuses Mercor of misappropriation of trade secrets and Ling of breaching his employment contract.

Scale claims Ling attempted to pitch Mercor to one of Scale’s largest clients—identified only in the suit as “Customer A”—before officially leaving. The company says the files include specific data that would let a competitor serve Customer A and other high-value accounts, potentially worth millions to Mercor if won.

Mercor denies wrongdoing, says it’s investigating

Mercor co-founder Surya Midha told TechCrunch the company has “no interest in any of Scale’s trade secrets” and is running the business differently. Midha acknowledged Ling may have had files stored in a personal Google Drive that Mercor has not accessed, and said the company offered to have those files destroyed or otherwise remediated.

Why this matters for the AI training market

The dispute exposes two fast-growing tensions in the AI data ecosystem: intense competition for enterprise customers, and the risk that human workflows—emails, drives, and departing employees—become vectors for transferring sensitive operational playbooks. Scale’s suit also arrives amid larger industry shifts, including Meta’s multibillion-dollar investment into Scale and a wave of talent movement between vendors.

For companies that prepare data to train large models, the stakes are practical and immediate: losing a major client to a rival using your internal labeling processes, customer mappings, or curated datasets can erase years of investment and revenue. Legal action is one remedy, but operational controls and rapid incident response are the first line of defense.

• Conduct exit interviews and immediate access revocation for departing staff to limit data exfiltration.
• Use file auditing and data-loss prevention (DLP) to track sensitive documents and detect unusual downloads or shares.
• Segment datasets and anonymize customer-specific mappings so a lost file can't be directly used to win a competitor’s contract.
• Prepare legal and forensic playbooks ahead of disputes to accelerate containment and preserve evidence.

Scale’s escalation to the courts signals that stakes have moved beyond contract clauses into litigation that can reshape vendor relationships. For clients, the episode is a reminder to ask suppliers about insider risk controls and to require clear data-handling assurances in contracts.

From a market perspective, rising startups like Mercor—who hire subject-matter experts to improve labeling quality—are changing competitive dynamics. That innovation creates value but also increases the need for stronger governance over how labeling knowledge and customer playbooks are stored and transferred.

QuarkyByte’s approach is to blend rapid investigation with practical hardening: map the data life cycle for your training pipelines, simulate the impact of leaked artifacts on revenue and customer relationships, and prioritize fixes that reduce risk fastest. Legal fights may follow, but prevention and fast response will limit damage and preserve customer trust.