-5.png?w=1024)
Quantum computers are notoriously difficult to operate. QuEra thinks Claude could make them a little less temperamental.
The quantum computing company said Anthropic's Claude developed and validated software capable of automatically recovering a critical laser system used to control its neutral-atom quantum computers. According to QuEra, the resulting system can recover from certain laser failures within seconds, compared with several minutes when a human specialist is required.
The experiment points toward a potentially important partnership between two emerging technologies: AI agents that can reason through complex engineering problems and quantum computers whose growing complexity increasingly demands sophisticated automation.
Claude learned how to recover QuEra's lasers
QuEra's quantum computers use neutral atoms as qubits and lasers to manipulate them. Those lasers have to remain at precise frequencies, but they can drift far enough that the system stops operating until someone corrects the problem.
Routine disturbances have already been automated. QuEra says its 256-qubit Aquila quantum computer, available through Amazon Braket, operates with uptime above 99%. More complicated laser failures, however, have traditionally required specialists who understand the system well enough to diagnose and recover it.
That is the problem QuEra gave Claude.
QuEra says a team of four specialists spent roughly two to three weeks creating an earlier recovery script that accounted for known failure conditions. Using a dedicated testbed, Claude performed experiments, evaluated the results, adjusted its approach and repeated the process across hundreds of failure cases.
In formal testing, QuEra said the resulting controller successfully recovered from 695 of 700 timed disturbances across seven fault classes. Most recoveries took less than six seconds, while the most difficult took roughly 10 to 14 seconds. QuEra said similar recoveries can take a human expert five to 10 minutes.
The work was conducted using Anthropic's new Model Hardware Standard, or MHS, a specification designed to enable AI agents to interact with physical scientific and manufacturing equipment.
Anthropic says MHS provides a standardized interface through which agents can read information from equipment and issue commands while remaining within predefined limits and safety controls. The research preview is also being tested with laboratory instruments, robotics, and manufacturing equipment.
Claude ultimately produced conventional, deterministic control software rather than becoming the AI that continuously controls the laser itself. The resulting program ran autonomously on the testbed, with QuEra now aiming to deploy the recovery system on live quantum processors.
QuEra engineers established the experiment's boundaries, supervised the process, and defined what constituted a successful result.
That distinction matters. The experiment is less about handing a quantum computer over to an autonomous AI than using AI to build automation for tasks that previously depended heavily on scarce human expertise.
Claude has increasingly been appearing in scientific work beyond traditional chatbot tasks. Anthropic recently said the model designed protein binders that were subsequently tested in a laboratory. Another Claude research model produced a new mathematical result while tackling a problem related to the Riemann hypothesis.
What eWeek found: Quantum's scaling problem isn't only about qubits
The eye-catching part of QuEra's experiment is the speed difference between automated laser recovery and manual intervention. The bigger story may be what happens as quantum computers become substantially more complicated.
Quantum machines require more than additional qubits to scale. They also need control electronics, lasers or cooling systems, calibration, error correction and software capable of keeping increasingly complex hardware operating reliably.
IBM's recent work on a massive modular quantum cooling system illustrates the same problem from a different direction. Building larger quantum computers increasingly looks like a systems-engineering challenge rather than a race centered exclusively on processor performance.
QuEra faces that problem particularly acutely because newer neutral-atom systems can require an increasing number of precisely controlled lasers. If every unusual failure requires one of a relatively small number of experts to intervene manually, deploying quantum computers outside tightly managed laboratories becomes harder.
AI-developed automation could help remove some of that bottleneck.
QuEra says it plans to explore using the same approach across additional quantum-computer subsystems. The company is also working toward Libra, a fault-tolerant quantum computer it plans to make available through Amazon Braket in 2028.
There are still important limits. Anthropic acknowledges that Claude's physical and spatial reasoning remains imperfect and that expert oversight remains necessary when agents interact with real-world equipment.
But QuEra's experiment suggests AI's contribution to quantum computing may arrive before AI begins running sophisticated quantum algorithms. Its more immediate role could be less glamorous and potentially just as important: helping engineers keep extraordinarily complicated quantum machines running in the first place.
Also read: Claude is pushing further into scientific research, with Anthropic saying its AI designed protein binders that were later validated in laboratory tests.