A control program written by Anthropic’s Claude restored a critical laser system on QuEra Computing’s quantum hardware in seconds, recovering from 695 of 700 simulated and real-world faults without a false success report, according to the Boston-based company.
The work was carried out through a research preview of the Model Hardware Standard, or MHS, a framework that lets AI agents operate lab and manufacturing equipment within predefined safety limits. The same results were discussed in a post on Anthropic’s blog.
The Laser Lock Problem
QuEra’s machines use precisely tuned lasers to manipulate neutral atoms that act as quantum bits, or qubits. The lasers must stay at exact frequencies, and small environmental changes can knock them off target, forcing operators to restore a so-called laser lock before any computation can continue.
QuEra has already automated recovery from routine laser disruptions, reporting uptime above 99% on its 256-qubit Aquila system, which is available through Amazon Braket. More complex failures, however, have still required a specialist to step in.
From Hand-Coded Recovery to Claude
The team started by handing the problem to four engineers, who spent two to three weeks writing a recovery script by hand. That script only covered the failure modes its authors had anticipated. QuEra then turned the job over to Claude through MHS. The framework let the agent run experiments on a dedicated laser testbed, study the outcomes and revise its approach. Engineers set the agent’s operating limits, reviewed its work and defined the conditions needed to demonstrate success. The hardware itself defined the safety boundaries, interlocks and emergency-stop procedures the agent was required to obey.
Claude did not stay in the loop once the program was finished. It produced a conventional piece of software that engineers could inspect and validate, and that program now handles recovery without any AI model making decisions at runtime.
Fault Trial Results
In 700 timed trials across seven fault types, the resulting controller brought the laser back to its target state 695 times and never reported a recovery that had not actually occurred. QuEra attributed the five failures to a problem in the testbed, not the controller. Most faults cleared in under six seconds, and the hardest took 10 to 14 seconds. QuEra said a specialist would normally need five to 10 minutes for the same job.
The testbed sat inside an active lab, so it dealt with normal foot traffic and environmental noise rather than only controlled conditions. QuEra said the controller recovered from every naturally occurring disruption during the pilot without human help.
Improved Lock Stability
The agent was also asked to improve the laser’s lock stability. QuEra reported that the new settings cut remaining noise by a factor of five and kept the system from losing its lock during unattended operation. An independent measuring instrument confirmed the settings matched an experienced specialist’s manual tuning while also fixing a problem the manual process had missed. When the same approach was applied to a laser running at a second wavelength, the agent worked out the settings in a single unattended overnight run, something QuEra said would normally take weeks of hands-on effort.
Why Self-Maintenance Matters
The practical motivation is straightforward. Neutral-atom machines are scaling up, newer systems use more lasers, and customer deployments can sit far from the engineers who built them. QuEra said manual tuning can take up to 30 minutes in some cases and may require someone on site at any hour.
“For years the hardest part of scaling quantum computers wasn’t the physics, it was the people driving at 2 am to fix a laser lock. We built a solution using the Model Hardware Standard to fix that: the lock recovers itself in seconds, verified every time, catching noise that’s easy to miss by hand. We’re building quantum computers that fix themselves,” said Sergio H. Cantu, vice president of quantum systems at QuEra.
Scope and Limitations
The pilot covered one subsystem. It does not show that an entire quantum computer can maintain itself, and QuEra did not report test results for other components. The company said the same development method could be extended to additional subsystems.
MHS began as a collaboration between Anthropic and the Howard Hughes Medical Institute’s Janelia Research Campus and remains in a limited research preview while its safety design is evaluated. Access is by application.
“We are among the best in the world at developing and operating quantum computers, and even for us, the cost of keeping these machines at peak performance is high,” said Takuya Kitagawa, president of QuEra. “A customer expects the entire computer, and thus every subsystem, to hold itself together without a specialist in the room. This is why the results from the MHS research preview and Anthropic’s frontier AI models are so meaningful. We are making it far easier and cheaper to keep our computers running at their best.”
QuEra is working with Amazon Web Services to deliver its planned Libra system through Amazon Braket in 2028, and has partnerships with Hewlett Packard Enterprise for on-premises high-performance computing integration and with NVIDIA for accelerated computing.