Fleet joins Anthropic project to test AI for security

Fleet has joined Anthropic's Project Glasswing, giving the device management company early access to Claude Mythos Preview for defensive security research.

It will use that access to examine how advanced AI models can help secure device management systems, mobile device management protocols and the infrastructure behind them. Planned tests include operational settings such as assessing endpoint state, setting remediation priorities and handling security risks more quickly.

The decision comes as security teams face a narrowing window between the disclosure of software flaws and their exploitation. Fleet cited its own research showing that 79% of organisations take more than a day to deploy critical security patches for browsers, operating systems and extensions, even as some attackers move within hours.

That gap has become a central issue for IT and security teams managing large numbers of corporate devices. Patching has long been a routine part of endpoint administration, but the pressure to make rapid changes safely across mixed fleets of laptops, phones and other connected systems has grown as attack methods have evolved.

Project Glasswing is Anthropic's controlled-access programme for Claude Mythos Preview. Through it, Fleet will test the model before wider release and assess whether it can improve internal security operations without adding new risks to sensitive environments.

The work will focus on defensive uses rather than broad automation claims. Areas under review include how AI can help teams understand the current condition of endpoints, identify the most urgent fixes and support faster responses when vulnerabilities or suspicious activity appear.

Fleet also linked the initiative to its open-source model and said it intends to publish lessons and practical guidance from the research. That would allow customers and other organisations working in device management and endpoint security to apply the findings in their own operations.

Outside experts say that approach matters as vendors come under pressure to react faster to AI-assisted attacks. The issue is not only whether suppliers can detect weaknesses quickly, but whether they can evaluate and act on them without causing disruption across large device estates.

"The speed of today's AI-driven attacks has fundamentally changed what organizations expect from their technology providers," said Gabe Chan, Staff IT Engineer at Abridge.

"Vulnerabilities are manipulated by bad actors almost instantaneously now. As a result, vendors must continuously improve how they identify, evaluate and respond to risk. We're encouraged to see Fleet investing in research that helps strengthen its own security while contributing practical learnings back to its customers and the industry," Chan said.

Security pressure

Endpoint management has grown more complex as companies spread work across office systems, remote laptops, mobile devices and cloud-connected assets. In that environment, a delayed patch or a missed signal from one part of the estate can create an opening that spreads beyond a single machine.

Fleet said the Glasswing work is intended to address that operational problem directly. Early access to the model allows the company to test emerging AI systems against real device management and security workloads rather than hypothetical scenarios.

That distinction matters because security teams have been cautious about giving AI systems too much autonomy. Tools that recommend a course of action may prove useful, but organisations still need confidence that changes will be accurate, auditable and safe before they are rolled out at scale.

Fleet manages devices across a wide range of platforms, including Macs, PCs, iPhones, iPads, Android devices, Chromebooks, Linux desktops, robotics and cloud infrastructure. That breadth gives it a varied testing ground for judging whether AI can assist defensive work across different operating environments.

Open approach

Mike McNeil, Chief Executive Officer and Co-Founder of Fleet, framed the move as part of the company's wider security efforts.

"As a company that specializes in making IT and security teams' lives easier, celebrating Fleet's founding anniversary with a renewed commitment to our own product security feels fitting," McNeil said.

"We are all about community and transparency. Since Fleet is open source, this Anthropic collaboration will help our customers and the wider security community understand where frontier AI can meaningfully improve the security of endpoint management workflows," he said.

Fleet's security leadership said the company intends to make the output public rather than keep it confined to internal development. That could make the programme relevant beyond Fleet's own products if the findings show where AI does and does not help defensive security work.

"As always, we'll share the lessons and best practices we develop in the open," said Dhruv Majumdar, Vice President, Security Solutions, Fleet.

"Everyone in the broader device management and endpoint security community can build on them," Majumdar said.