The Perseverance rover from NASA has marked several groundbreaking achievements since its arrival in 2021, such as transmitting the initial sound clips from the Martian landscape. Almost five years following its touchdown on the planet, the rover has now accomplished yet another milestone. In December, it traversed a segment of the Jezero crater following a path designed by Anthropic's Claude AI model, representing the inaugural instance of NASA employing a large language model to guide the vehicle-sized explorer.
From December 8 to 10, the rover traveled roughly 400 meters, equivalent to around 437 yards, across a rocky terrain on Mars that Claude had outlined. Employing an artificial intelligence system to chart the rover's trajectory involved more than just entering a basic instruction.
According to NASA, directing the Perseverance rover presents significant challenges, regardless of whether done by people or machines. The agency noted that each journey requires meticulous preparation to avoid risks like slipping, overturning, wheel spinning, or becoming stuck. Since its deployment, the rover's team has meticulously defined navigation points—referred to as a 'breadcrumb trail'—drawing on orbital imagery alongside the vehicle's internal cameras.
For Claude to undertake the assignment, NASA initially supplied Anthropic's Claude Code, a coding tool, with extensive historical information gathered by the rover over years, enabling the AI to start generating a navigation plan. The model proceeded systematically, assembling navigation points across 10-meter intervals, which it subsequently reviewed and refined.
Given the rigorous standards at NASA, specialists at the Jet Propulsion Laboratory (JPL) thoroughly verified the AI's output prior to uploading it to the rover. The JPL staff tested Claude's navigation points via their routine simulation software, which validates instructions for the vehicle. Ultimately, the agency reported needing only slight adjustments to the proposed path, including one modification based on on-site photos unavailable to Claude during its analysis.
NASA's engineers project that integrating Claude will reduce navigation planning duration by 50 percent and enhance the uniformity of the rover's travels. With reduced hours on laborious hand-crafted routes and operator training, the team can execute additional excursions, gather greater volumes of research material, and perform more in-depth examinations. Overall, this approach promises deeper insights into the Martian environment.
Although AI's efficiency improvements are sometimes exaggerated, for NASA, any resource boosting staff productivity is highly valued. This past summer, the organization shed around 4,000 positions, roughly 20 percent of its total staff, amid reductions imposed by the Trump administration. As 2026 approached, the president's budget proposal sought to slash the science funding by almost 50 percent, a measure Congress blocked in early January. Nevertheless, with resources maintained slightly under 2025 figures, NASA faces considerable obstacles. The agency must resume lunar missions with fewer than half the personnel available at the Apollo era's peak.
This accomplishment stands as a significant advancement for Anthropic. Observers might remember that Claude struggled to complete Pokémon Red last spring. Within under 12 months, Anthropic's AI has progressed from difficulties with a basic 8-bit handheld game to effectively designing a trajectory for an extraterrestrial rover. NASA anticipates ongoing partnerships, highlighting that self-governing AI could assist spacecraft in venturing farther into the solar system.