The Simple Version — What Mercor Actually Does
Think about how an AI like ChatGPT learns. It does not just read Wikipedia and call it a day. To become genuinely useful — to answer a legal question properly, to analyse a medical case correctly, to review code like a senior developer — an AI needs to be trained by people who actually know those things.
That is exactly what Mercor does. It finds highly qualified professionals — former Goldman Sachs bankers, experienced doctors, senior lawyers, seasoned engineers — and pays them to work on AI training tasks. These experts might review AI-generated answers and mark them right or wrong. They might demonstrate how they would solve a problem in their field. They might have conversations with AI systems and correct mistakes.
The AI companies — OpenAI, Anthropic, Meta, Google — pay Mercor for access to this expert knowledge. Mercor takes a cut and passes the rest to the experts. It is essentially a marketplace that connects the people who know things with the machines that are trying to learn those things.
Mercor is the company that hires smart people to teach smart machines — acting as the middleman between human expertise and AI training.
By the Numbers — How Big Is Mercor?
Who Founded Mercor and How Did It Start?
Mercor was founded in 2023 by three friends from the San Francisco Bay Area — Brendan Foody, Adarsh Hiremath, and Surya Midha — who had competed together on their high school debate team. They all dropped out of college to start the company and were later awarded Thiel Fellowships, a programme that pays young people $100,000 to skip university and build something instead.
The company originally started as a platform to connect Indian software developers with US companies looking to hire. They built AI tools to interview and match candidates — and then realised that the real opportunity was not recruitment, it was using that same idea to connect expert professionals with AI companies that needed high-quality training data.
That pivot made them three of the youngest self-made billionaires in history. In October 2025, Bloomberg named them among the 24 AI startups to watch in 2026. They were 22 years old at the time.
What Kind of Work Do Experts Do on Mercor?
The work varies depending on what the AI company needs. But in general, experts on Mercor might be asked to:
- Review answers that an AI gave to legal, medical, or financial questions — and rate them as correct, partially correct, or wrong
- Demonstrate how they would approach a problem in their field — showing the AI what good thinking looks like
- Have back-and-forth conversations with an AI to help it learn how real professionals communicate
- Test whether an AI can complete real-world tasks in their domain — and flag where it fails
The pay is genuinely good. Top specialists — particularly in medicine, law, and finance — can earn up to $250 to $300 per hour. More typical work pays less, but for professionals who can do this work flexibly from home, it is an attractive income source.
There is something philosophically interesting about what Mercor does. Professionals are being paid good money to teach AI systems how to do their jobs. Some of that AI training will eventually make parts of their work automatable. They are, in a sense, training their own replacement — while getting paid well to do it. Mercor's CEO Brendan Foody believes this is inevitable: he thinks eventually all knowledge work will shift toward training AI agents rather than doing the underlying work directly. That is a big claim. Whether it is right, we will find out over the next decade.
Mercor's Customers — Who Uses It?
Mercor's customers are the biggest names in AI. According to the company, it works with the top five AI labs and six of the "Magnificent Seven" tech companies. That means its training data and expert evaluations are shaping the AI systems that hundreds of millions of people use every day — ChatGPT, Claude, Gemini, and others.
The company also publishes its own AI benchmarks — called APEX — which measure how well frontier AI models perform at real professional tasks in fields like investment banking, corporate law, and software engineering. These benchmarks are independent evaluations, not advertisements, and are taken seriously by the AI research community.
The Data Breach — What Happened in March 2026?
In late March 2026, Mercor was caught in a major cybersecurity incident. Hackers planted malicious code inside LiteLLM — an open-source software tool used by thousands of AI companies to connect their applications to AI services. The code was designed to steal credentials and data before it was identified and removed.
Mercor confirmed it was one of thousands of companies affected. But because of what Mercor holds — the personal information of 40,000+ contractors including government ID documents, Social Security numbers, and video interview recordings — the impact was more severe than for most other companies hit.
A hacking group called Lapsus$ later claimed it had accessed Mercor's data, posting samples including Slack conversations and internal videos. Meta indefinitely suspended all work with Mercor following the breach. Class action lawsuits were filed representing the affected contractors. The incident was a significant blow to a company whose entire value proposition is being a trusted middleman for sensitive AI training data.
If you are considering working with Mercor, be aware that the March 2026 breach exposed contractor PII including identity documents. Mercor says it has improved security since the incident, but it is worth understanding what data the platform collects before signing up. Read their data usage policy carefully at talent.docs.mercor.com.
Should You Work for Mercor?
If you are a qualified professional — a doctor, lawyer, engineer, banker, or journalist — and you are looking for flexible extra income, Mercor is worth looking at. The pay is genuinely good, the work is flexible, and you are contributing to something that matters. The March 2026 breach is a real concern, and you should read the company's current security disclosures before sharing personal information.
If you are a company looking to build better AI, Mercor's expert-driven approach is a meaningful alternative to cheaper, lower-quality crowdsourced training data platforms. The benchmarks the company publishes suggest its approach to data quality is serious, not just marketing.
What Mercor has built is genuinely important for AI development. The difference between an AI that gives you generic advice and one that can reason like a domain expert is almost entirely a function of training data quality. Mercor's bet — that expert human knowledge is the key ingredient — is correct. What I find most interesting is the benchmark they publish. Companies that publish their own independent benchmarks and invite scrutiny are usually more confident in their quality than those who do not. The breach was damaging and the lawsuits are real. But the underlying business model is sound, and the demand for what they do is only going to grow.
Frequently Asked Questions
Is Mercor AI legit?
Yes — Mercor is a real company with real customers including OpenAI and Anthropic. It is valued at $10 billion and has raised $350 million in funding from credible venture capital firms including Felicis Ventures. It is not a scam or a gig-work platform in the traditional sense — it specifically targets credentialed professionals in specialised fields.
How much does Mercor pay?
Pay rates on Mercor vary by specialisation and task type. The platform targets professionals in high-value fields and pay rates for top specialists in law, medicine, and finance can reach $250 to $300 per hour according to multiple reports. More typical work pays less. Mercor earns a margin on top of the rates it charges to AI companies.
What happened to Mercor after the data breach?
Following the March 2026 breach, Meta indefinitely suspended its work with Mercor, class action lawsuits were filed representing over 40,000 affected contractors, and the company faced significant reputational damage. Mercor confirmed it was affected by the LiteLLM supply-chain attack and said it was conducting forensic investigations. The company's long-term partnerships with other AI labs appear to remain intact as of May 2026.
How is Mercor different from Scale AI?
Both Mercor and Scale AI provide training data and human evaluation for AI companies. The key difference is who does the work. Scale AI primarily uses crowdsourced labour — large numbers of general workers completing relatively simple tasks. Mercor specifically recruits credentialed professionals from elite institutions to work on complex, knowledge-intensive tasks. Mercor's CEO argues that as AI gets better at simple tasks, the value shifts entirely to expert-level training data — which is Mercor's core focus.