PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteAI labs need professional expertise to build and test systems that can do expert work, but companies may be unwilling or unable to license their internal data. Mercor offers another route: it recruits specialists to create examples, judge model answers, and demonstrate workflows. That is not the same as acquiring a company’s database—and it raises hard questions about confidentiality, worker rights, and security.
What Mercor does—and what it does not necessarily obtain
Mercor is an intermediary connecting AI companies with people who have professional expertise. Those experts can write answers, review model outputs, create scoring rubrics, correct errors, and work through tasks that call for domain judgment. The resulting material can help train models or test whether they perform reliably.
![Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results](https://m.media-amazon.com/images/I/415+fSJacsL._SL160_.jpg)
The distinction matters: Mercor’s core model is not simply to obtain confidential corporate databases. It turns some professional knowledge and judgment into structured work products. Separately, Mercor markets enterprise data services, including a process it says can anonymize operational data before delivery to AI labs. That is a company representation, not independent proof that re-identification is impossible or every underlying permission has been secured. Mercor’s data-services description says lab buyers do not see raw data.
TechCrunch reported in October 2025 that Mercor’s customers included OpenAI, Anthropic, and Meta. That reporting describes customer relationships at that time, not necessarily current contracts. TechCrunch’s report on Mercor and AI-lab data also said some assignments paid as much as $200 an hour.
#1 Best Overall
Why AI labs pay for expert judgment
Public web text is abundant, but abundance is not the same as a dependable standard of professional quality. A model may produce a plausible legal analysis while missing a material issue, or generate working-looking code that is unsafe in production. To improve and assess these systems, developers need examples and judgments that distinguish plausible from professionally defensible.
Mercor describes work that includes creating examples, reviewing model responses, and assessing outputs against professional standards. Its software-engineering work, for example, can involve debugging, code review, architecture, distributed systems, and production failures. Mercor’s expert program and software-engineering work description outline these categories.
- Training material can include demonstrations, corrections, explanations, preference judgments, and task solutions used to shape a model’s behavior.
- Evaluation material can include held-out tasks, benchmark questions, rubrics, expert scores, and simulated workflows used to test a model or agent.
Those are different products. Training material teaches or reinforces behavior; evaluation material measures performance. Evaluation sets can be especially sensitive: if a model is trained on the test tasks, apparent improvement may reflect memorization rather than broader capability. A benchmark can also reveal what a lab considers important or where its systems are weak.
Why a company may not license its internal data
A company considering a deal with an AI lab faces more than a pricing question. Its documents and systems may contain customer information, trade secrets, privileged material, source code, regulated data, or work product whose use is restricted by contracts. It may also be unclear whether the company has authority to license material created by employees or derived from customers.
Free tools Windows power users keep installed
One-click scans. No signup required.
- Competitive incentives: A company may not want to help train a system that could automate services it sells.
- Legal and governance exposure: Privacy, copyright, employment, professional-responsibility, and contractual obligations can make data preparation and licensing complicated.
- Strategic choice: A business may prefer to develop its own AI capabilities rather than strengthen a potential competitor.
- Operational burden: Cleaning, labeling, anonymizing, documenting, and approving internal data can require substantial work.
Mercor CEO Brendan Foody has argued that firms such as investment banks may resist giving AI labs the information needed to automate their value chains. Goldman Sachs is an example in that argument, not evidence that Goldman Sachs provided data to Mercor or participated in its projects. TechCrunch reported Foody’s argument in that context.
How the expert-work pipeline works
Mercor describes a process in which a professional creates a profile, completes an adaptive assessment or interview, and may be matched to projects based on expertise. Project work can involve reviewing or creating material for AI systems. Mercor says projects may last weeks or months and can be extended, shortened, or ended early; compensation depends on the assignment, expertise, complexity, demand, and project structure.
- Apply: Create a professional profile through Mercor’s expert page.
- Demonstrate fit: Complete the assessment or interview used to evaluate relevant skills.
- Review the assignment: Check the project scope, rate, duration, confidentiality terms, ownership terms, and any restrictions before accepting.
- Complete the work: Produce examples, assess outputs, or follow the project’s evaluation rubric.
Marketplace listings observed in August 2026 included some physician and machine-learning engineer roles at approximately $110–$250 an hour, lawyers at $60–$150, financial analysts at $60–$180, and writing specialists at $75–$100. These are project listings, not guaranteed rates or a universal pay schedule; location, credentials, demand, and assignment terms may change what an individual is offered. Mercor’s marketplace is the source for those listings.
Mercor’s figures about scale also need attribution. Its company materials report more than 5 million experts, more than 200,000 trained evaluation authors, and coverage across 300-plus professional domains; these are company-reported figures, not independently audited counts. Mercor’s enterprise-evaluations page and newsroom present its current scale claims.
Recommended Free Tools
The boundary between expertise and company information
Foody has argued that knowledge in an employee’s head belongs to that employee. That is a business position, not a universal legal rule. General skills and experience can be distinct from protected information, but the boundary depends on the information, the worker’s agreements, the circumstances, and applicable law.
- Generally lower risk: Broad professional skills, publicly known methods, personal judgment, and experience that is not tied to an employer’s confidential materials.
- Higher risk: Trade secrets, customer-specific details, internal pricing or investment models, privileged legal information, source code, internal prompts or rubrics, and information subject to an NDA or employment agreement.
Consider a former employee asked to explain a familiar workflow. A general description may draw on transferable experience; reproducing the employer’s exact internal template or a client-specific method may cross into protected material. A current employee taking side work in the same field faces additional conflict-of-interest and contract questions. A synthetic case study based on a real client matter can still expose identifying or confidential details.
Rank #3
Mercor’s legal support materials say contractors can access documents including Terms of Work and a Confidential Information and Inventions Assignment Agreement. Its social-media policy prohibits sharing client names, screenshots, detailed task content, or confidential material. Those policies do not settle every worker’s legal obligations: anyone considering a project should read the actual terms and existing employment or client agreements. Mercor’s legal-support materials and social-media policy describe its rules.
Why access to a production codebase is a sharper test
TechCrunch reported that one Mercor job posting sought a startup CTO or co-founder able to authorize access to a substantial production codebase for AI evaluations or possible training. Mercor told the publication that some startup CTOs accepted such offers, but did not provide contract details. This is a reported example, not proof that Mercor routinely obtains unauthorized code.
A person who can access a codebase may not have authority to license it. The relevant questions include who owns the code, whether the person has corporate approval, what the agreement permits, and whether customer or employee data is embedded in the repository.
How Mercor makes money, and how its business has broadened
The basic economics are those of a specialized intermediary: experts are paid for work, while the buyer pays for recruitment, matching, project management, quality control, and delivery as well as the work itself. High-value expert judgment may be more useful to a lab than large volumes of generic annotation, but the value depends on quality, provenance, and the rights the buyer receives.
TechCrunch reported in October 2025 that Mercor claimed tens of thousands of contractors, more than $1.5 million in daily contractor payments, and roughly $500 million in annualized recurring revenue. Mercor’s later newsroom materials reported $4 million paid to its network daily, more than 400 employees, and a $10 billion valuation. These are figures from different dates and company or publication reporting; they should not be treated as audited financial statements or combined as if measured on the same basis. TechCrunch’s October 2025 report and Mercor’s newsroom provide the respective figures.
Rank #4
Mercor now markets beyond expert labor for frontier labs. Its enterprise-evaluation offering includes expert staffing, managed task-based evaluation, and self-serve evaluation tools for comparing models, tools, context, cost, and post-training returns. The company says a reusable evaluation suite can take roughly four to six weeks to build; that is its estimate, not a guaranteed delivery time. Mercor’s enterprise-evals page describes those offerings.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
It also markets workflow diagnostics, agent deployment, benchmarking, and enterprise data monetization. A company considering these services needs to establish data authority and consent before sharing operational information, including whether employee and customer permissions cover the proposed use. Mercor’s enterprise page describes its agent and workflow services, while its data page describes data monetization.
Competition and the cost of disputes
Mercor competes in a market that includes large data-labeling companies, specialist expert-data providers, evaluation firms, staffing marketplaces, direct enterprise licensing, and in-house AI teams. TechCrunch has identified Scale AI and Surge among providers moving toward expert data and AI-agent environments.
Scale AI sued Mercor and a former Scale employee, alleging trade-secret misappropriation and breach of contract. Those are allegations in litigation, not findings of fact. The case illustrates the commercial stakes in customer relationships and know-how, but does not by itself establish wrongdoing by Mercor. TechCrunch’s account of the lawsuit and Axios’s report describe the claims.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.The 2026 security incident changed the risk calculation
Mercor disclosed that it was affected in late March 2026 by compromised versions of the LiteLLM open-source package, which were designed to exfiltrate credentials. The company says it contained unauthorized activity and investigated with Mandiant, Latacora, law enforcement, and industry partners; it says its investigation found employee data was not affected. Mercor’s incident update sets out its account.
Best Value
WIRED reported that Meta paused work with Mercor while investigating and that other AI labs reassessed their relationships. That reporting supports a pause during review, not a claim that Meta permanently ended the relationship. WIRED’s report describes the customer response.
Forbes reported former-employee concerns involving operational failures, suspected fraud, and possible North Korean infiltration. Those reports and allegations should not be treated as official findings. The public accounts cited here do not establish whether contractor identities, recordings, prompts, rubrics, client instructions, or training data were accessed or exfiltrated; nor do they establish the incident’s full scope or the duration of any customer pauses. Forbes’s report attributes the concerns it describes.
The security stakes extend beyond the underlying training examples. A platform may hold contractor identities and résumés, interview recordings, customer instructions, project metadata, evaluation rubrics, and credentials. Even metadata can reveal what a lab is testing or building. For a buyer, the vendor’s security controls, access segmentation, incident notification, retention, and independent audit rights are part of the data decision—not a separate technical detail.
What buyers and professionals should check
For an enterprise buyer
- Confirm ownership, license scope, exclusivity, and chain of title for deliverables.
- Ask how contributors are screened for conflicts, current employment, and relevant confidentiality obligations.
- Establish whether customer data, prompts, or evaluation sets can be used to improve Mercor’s own systems or shared with other clients.
- Specify project isolation, subprocessors, retention and deletion, breach notification, audit rights, and data residency.
- Keep evaluation tasks separate from training material and define controls against benchmark contamination.
- For operational data, document the basis for employee and customer consent and verify the anonymization process rather than relying on the label alone.
For a professional considering contract work
- Read the project terms, confidentiality obligations, and invention-assignment language before accepting.
- Do not use an employer’s documents, code, client details, templates, or nonpublic methods unless you have clear authorization.
- Ask who owns your work, how long it may be reused, and whether interview recordings or other personal data are retained or used for platform purposes.
- Clarify payment timing, project duration, required monitoring, and what happens if a project ends early.
- Do not assume that a contractor label resolves worker-classification questions in every jurisdiction.
Mercor’s privacy policy says it collects information such as resumes, work histories, interview recordings and transcriptions, profile photos, salary expectations, device data, and usage information. It also says profiles, resumes, and interview data may be shared with companies using the platform and that third-party services may be used for AI evaluation and image processing. Read the policy and project-specific terms before submitting sensitive information. Mercor’s data privacy policy describes its stated practices.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
