Anthropic keeps making its hiring tests harder because candidates cheat with Claude

ANTHROPIC KEEPS MAKING ITS HIRING TESTS HARDER BECAUSE CANDIDATES CHEAT WITH CLAUDE

Powerful language models are now in the hands of a broad user base, and that puts traditional ways of assessing professional skills under unprecedented pressure. The performance optimization team at Anthropic, the US company behind the frontier model Claude, has hit an unusual problem: its own creation — or, more precisely, AI tools like it — has started "cheating" in job interviews.

Since early 2024, Anthropic has used take-home assignments to test the technical skills of engineering candidates. The format is meant to let applicants show what they know in a calmer, more considered setting than a standard interview allows. But AI coding tools such as GitHub Copilot, along with steadily improving large language models, have made these assignments too easy to solve with AI help. That made it hard to assess candidates objectively, because the company could no longer say with confidence who had actually solved the task — the applicant or the AI.

To push back and prevent cheating, Anthropic's engineers have to keep revising the assignments and making them harder. The goal is to write problems that demand deep contextual understanding, unconventional thinking and creativity — areas where today's AI models, impressive as they are, still trail humans. These are not tasks that ask candidates to reproduce well-known algorithms or write boilerplate code. Instead, they require analyzing complex systems, making decisions under uncertainty, connecting scattered pieces of knowledge and showing original thinking. Such problems are harder to automate and harder to "feed" to a language model for a ready-made answer.

The situation highlights how difficult it is becoming to assess human skills when powerful AI tools are available everywhere. The line between human competence and AI capability keeps getting blurrier. Companies around the world are starting to look at new approaches to recruiting and evaluating staff that reflect today's technology landscape.

The future may lie in evaluating not just the final result but the problem-solving process itself: tracing a candidate's reasoning and testing their ability to adapt and learn. By running into this problem at the cutting edge, Anthropic effectively shows how the industry is being forced to adapt to a world where AI is not only a work tool but also a factor that rewrites the rules of hiring.

In the end, Anthropic's constant toughening of its take-home tests is a clear example of how technological progress demands flexibility and new ideas in every area of human activity, hiring included. It forces companies to rethink what professional fitness means and to look for new, more reliable assessment methods that can tell genuine human talent apart from skillfully generated AI answers. In an era when AI is becoming a universal assistant, we need new ways to judge what it actually means to be a competent professional.