An independent study by nonprofit METR found Anthropic is developing artificial intelligence nearly twice as fast as before. Researchers estimated that internal artificial intelligence tools drove 80 percent of this acceleration, raising safety concerns about automated self-improvement.