Anthropic Nabs Vercept for Claude's OS Skills

Anthropic has acquired Vercept, a startup specializing in AI perception and interaction, to significantly advance Claude's ability to operate within live computer applications. The move aims to empower Claude to execute complex, multi-step tasks directly within software, mirroring human interaction.

Vercept's team, including co-founders Kiana Ehsani, Luca Weihs, and Ross Girshick, built their company on the thesis that effective AI requires solving difficult perception and interaction challenges. Their expertise in enabling AI systems to "see and act" within everyday software directly addresses Anthropic's goals for Claude's expanded functionality. Vercept will discontinue its external product and integrate its operations into Anthropic.

This acquisition follows recent strides in Claude's "computer use" capabilities. The newly launched Claude Sonnet 4.6 model achieved a 72.5% score on OSWorld, a key evaluation benchmark for AI computer use, a substantial leap from under 15% in late 2024. Sonnet 4.6 is now nearing human-level performance on tasks like spreadsheet navigation and web form completion.

Anthropic seeks teams whose technical ambitions align with its own, emphasizing advancements grounded in safety and rigor. Vercept's founding mission—to build AI that works directly on a user's computer, understanding screens and workflows safely—resonates with Anthropic's commitment to responsible AI development. This marks Anthropic's latest acquisition, following its earlier integration of Bun.