Mercor recently published a benchmark evaluating AI agents' performance on professional tasks, including law and corporate analysis. Initial scores were low, with major labs failing to exceed 25%, suggesting that lawyers were safe from AI displacement for the time being. However, the landscape shifted with the release of Anthropic's Opus 4.6, which improved AI performance significantly. The new model scored nearly 30% in one-shot trials and averaged 45% when given multiple attempts, marking a notable advancement in AI capabilities. The jump from an 18.4% score to 29.8% in just a few months indicates rapid progress in foundation models. This improvement is partly attributed to new features like "agent swarms," which enhance the AI's ability to tackle complex, multistep problems. Mercor's CEO, Brendan Foody, expressed his astonishment at the speed of this development. While a score of 30% is still far from perfection, it raises concerns for the legal profession, suggesting that lawyers should remain cautious about their job security in light of these advancements.