Rockwell Automation is joining Anthropic’s Project Glasswing to test advanced AI for defensive cybersecurity across industrial automation and connected manufacturing systems. The collaboration gives Rockwell security teams controlled access to Claude Mythos 5, Anthropic’s restricted cybersecurity model, to investigate how AI can accelerate vulnerability discovery, validation, prioritization and remediation in critical industrial software.
Rockwell Automation is bringing industrial automation expertise to Anthropic’s Project Glasswing, a cybersecurity initiative built around using highly capable AI models to identify and remediate vulnerabilities in critical software.
The collaboration gives Rockwell’s security teams controlled access to Claude Mythos 5, the latest Mythos-class model, for defensive cybersecurity work. Rockwell says it will explore the model’s ability to accelerate vulnerability discovery, validation, prioritization and remediation across software and connected systems used in manufacturing.
The partnership places AI-driven vulnerability research directly inside the operational technology (OT) environment, where software flaws can have consequences beyond conventional IT systems. Industrial control systems, automation platforms and connected production infrastructure can influence physical processes, making cybersecurity an operational-resilience issue as well as an information-security concern.
Rockwell’s participation is therefore significant because it extends Project Glasswing into the industrial automation ecosystem. The company supplies technologies used across manufacturing environments, while Anthropic is providing access to a model specifically restricted for defensive cybersecurity applications.
Project Glasswing launched in April 2026 with a group of technology and infrastructure organizations, including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks. Anthropic subsequently expanded the program to approximately 150 additional organizations across more than 15 countries.
Anthropic’s rationale for the initiative is based on the growing capability of frontier AI models to find software vulnerabilities. In an update published in May, the company said roughly 50 initial Project Glasswing partners had used Claude Mythos Preview to identify more than 10,000 high- or critical-severity vulnerabilities.
Anthropic has since moved the program from Mythos Preview to Claude Mythos 5. The company says Mythos 5 is available only to a small group of vetted partners because of its cybersecurity capabilities. Its current Mythos 5.1 model is similarly restricted through trusted access programs.
For Rockwell, the immediate objective is not to replace existing security processes with autonomous AI. Instead, the company says it is investigating how advanced AI can strengthen the security lifecycle by helping teams identify and address weaknesses more quickly.
That distinction is important in industrial environments. Vulnerability management in OT often has to account for production availability, equipment lifecycles, safety requirements and dependencies between software and physical systems. A vulnerability that can be patched immediately in an ordinary IT environment may require considerably more assessment in a manufacturing operation.
Rockwell’s approach also reflects the growing use of AI for cybersecurity, particularly for tasks involving large codebases and complex software dependencies. Anthropic says Mythos Preview was able to identify complex vulnerabilities and, in some cases, develop proof-of-concept exploits during controlled testing. That capability is one reason the model has remained gated rather than becoming a general-purpose public model.
The defensive opportunity is accompanied by a corresponding security challenge. If AI can identify vulnerabilities faster, defenders may need to shorten the time between discovery, validation, disclosure and patching. Anthropic said its early Project Glasswing experience shifted the bottleneck from discovering vulnerabilities toward verifying, disclosing and patching the large numbers of flaws identified by AI.
That dynamic is particularly relevant to manufacturers with large portfolios of connected industrial products.
Gartner estimates worldwide information-security spending will reach $244 billion in 2026, representing 11.6% constant-currency growth. The analyst firm says organizations are increasing security investment to address threats and improve resilience, while cybersecurity products are expected to provide AI-driven benefits and secure AI usage.
Gartner is also tracking the specific risks associated with AI in operational environments. The firm predicts that by 2028, misconfigured AI in cyber-physical systems could shut down national critical infrastructure in a G20 country. Its definition of cyber-physical systems includes operational technology, industrial control systems, industrial automation, industrial IoT and robotics.
The convergence of these trends creates a two-sided requirement for industrial companies: AI must become part of the cybersecurity response while the AI itself must be governed carefully.
Rockwell’s Project Glasswing participation fits that model. The company can use advanced AI to examine the software underlying connected industrial environments while retaining security teams as the decision-makers responsible for validating findings and determining remediation.
The broader market is moving in the same direction. Microsoft, Google, NVIDIA, CrowdStrike and other technology companies are investing in AI-assisted cybersecurity, while industrial vendors are increasingly incorporating security capabilities into automation and connected-device platforms.
For Rockwell, the partnership extends its secure-by-design strategy into a new generation of AI-assisted vulnerability research. For Anthropic, the collaboration provides access to an industrial technology environment where software security is directly connected to manufacturing operations.
The longer-term question will be whether highly capable AI can reduce the time required to secure industrial software without introducing new operational risks. Project Glasswing is effectively being used as a controlled testing ground for that transition, with Rockwell now adding the manufacturing and industrial automation perspective.
Market Landscape
AI-assisted cybersecurity is moving from automated alerting toward vulnerability discovery, code analysis, remediation and continuous defensive testing. Project Glasswing represents one of the more tightly controlled examples, restricting access to Anthropic’s Mythos models to vetted organizations conducting defensive work.
The industrial sector presents a particularly important application because OT environments combine software, networking and physical processes. Gartner says cyber-physical systems—including industrial automation and control systems—face emerging risks from misconfigured AI.
Meanwhile, Gartner forecasts global information-security spending of $244 billion in 2026. Its research also identifies AI cybersecurity as a rapidly expanding category, with AI cybersecurity spending forecast at $51.3 billion in 2026.
The competitive landscape includes cybersecurity vendors, hyperscalers, AI model developers and industrial technology companies. The emerging focus is not simply whether AI can find vulnerabilities, but whether organizations can safely integrate those capabilities into existing security, development and operational workflows.
Top Insights
- Rockwell Automation is testing Claude Mythos 5 for defensive vulnerability discovery and remediation across connected industrial software.
- Project Glasswing restricts access to Anthropic’s highly capable cybersecurity models to vetted organizations conducting defensive security work.
- AI-assisted vulnerability research could help industrial companies shorten the time between identifying software flaws and addressing them.
- Industrial AI security requires controls that account for operational technology, connected systems, production availability and physical-world consequences.
- Anthropic’s early Project Glasswing results highlight both the defensive potential and governance challenges of highly capable cybersecurity AI.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
