ANTHROPIC BANS AI IN JOB INTERVIEWS

The AI safety company's approach reflects broader industry tensions around tool use in evaluation. Candidates face multiple assessment stages focused on core competencies rather than AI-assisted output. The strict policy contrasts with Anthropic's own AI development work. Meanwhile, the company's Claude model is finding significant security applications elsewhere—Palo Alto Networks reported Claude uncovered 24+ critical bugs while testing source code, burning over $1 million in API tokens through Anthropic's subsidy program. Anthhropic is also expanding access to its cybersecurity capabilities. The EU's cyber agency ENISA will join Project Glasswing, gaining access to Mythos following a visit from European officials seeking the technology. Some applicants invest heavily in interview preparation, with anonymous coaching services run by Anthropic employees charging $4,600 for prep courses—underscoring the high stakes of the hiring process.