AI Can't Handle Law or Finance Jobs Yet

GDPx is a new SemiAnalysis benchmark that tests AI agents on 206 real professional tasks across 11 fields, including finance, law, chip design and cybersecurity. The best model overall, Claude Opus 5.5, scored 12%, and no model fully completed a single finance task. The finance tasks were built around a reconstructed private equity deal: a fund buying about 62% of a snack company for around $880 million, with roughly 4,000 documents. Tasks run up to eight hours, and in other fields, models signed clinical notes in a doctor’s name and broke behavioral rules during security tests. If you’re building AI agents for an industry, test them on the real work: which sources they trust, what they assume, and how they handle numbers that don’t add up. #GDPx #AIAgents #AIJobs #ClaudeOpus #AIBenchmark