Is OpenAI Crossing Ethical Lines to Train Its AI?
In a move that’s raising eyebrows across the tech industry, OpenAI is reportedly asking third-party contractors to upload real work assignments from their current or past jobs. The goal? To evaluate the performance of its next-generation AI models against human professionals. But here's where it gets controversial: this approach could potentially expose sensitive corporate data and trade secrets, sparking ethical and legal concerns. According to records obtained by WIRED, OpenAI, in collaboration with training data company Handshake AI, is collecting real-world tasks to establish a human baseline for comparison with AI models. And this is the part most people miss: the company is specifically requesting concrete outputs—think Word docs, PDFs, or Excel files—rather than summaries, to ensure the data reflects authentic, on-the-job work.
This initiative is part of OpenAI’s broader effort to measure its progress toward achieving Artificial General Intelligence (AGI), an AI system capable of outperforming humans at most economically valuable tasks. In September, the company launched a new evaluation process to benchmark its AI models across various industries. A confidential OpenAI document explains, ‘We’ve hired folks across occupations to help collect real-world tasks modeled off those you’ve done in your full-time jobs, so we can measure how well AI models perform on those tasks.’ Contractors are instructed to take long-term or complex projects they’ve completed and break them down into specific tasks, ensuring the examples are ‘real, on-the-job work’ they’ve ‘actually done.’
For instance, one example in an OpenAI presentation describes a task for a Senior Lifestyle Manager at a luxury concierge company: creating a 2-page PDF draft of a 7-day yacht trip itinerary to the Bahamas for a high-net-worth family. The contractor would upload an actual itinerary they’ve previously created for a client, minus any confidential details. OpenAI emphasizes the need to scrub files of corporate intellectual property and personally identifiable information, even providing tools like ‘Superstar Scrubbing’ to guide contractors. But is this enough to avoid legal pitfalls?
Here’s the controversial question: Are AI labs like OpenAI putting themselves—and their contractors—at risk by relying on individuals to determine what constitutes confidential information? Intellectual property lawyer Evan Brown warns that contractors could inadvertently violate nondisclosure agreements or expose trade secrets, leaving both parties vulnerable to legal claims. ‘The AI lab is putting a lot of trust in its contractors,’ Brown notes. ‘If they do let something slip through, are the AI labs really taking the time to determine what is and isn’t a trade secret?’
This strategy isn’t unique to OpenAI. Companies like Anthropic and Google are also hiring vast networks of contractors to generate high-quality training data, creating a lucrative sub-industry within AI development. Handshake AI, for example, was valued at $3.5 billion in 2022, while Surge AI reportedly sought a $25 billion valuation in recent funding talks. OpenAI has even explored acquiring data from companies that have gone out of business, though concerns about fully anonymizing personal information reportedly stalled such efforts.
While the push for real-world training data is driving AI innovation, it raises critical questions about ethics, privacy, and legal boundaries. What do you think? Is OpenAI’s approach a necessary step toward advancing AI, or does it cross a line? Share your thoughts in the comments—let’s spark a conversation about where we draw the line in the pursuit of technological progress.
