Secure AI starts with resilient data

Enterprise AI does not operate in isolation. It retrieves, interprets, and increasingly acts on company data. If that data is exposed, corrupted, or unavailable, the impact can extend well beyond a bad answer. Agents can make the wrong decision, business processes can stop, and recovery can become far more complicated.

That is the perspective Cohesity is bringing as we sign on to both the Open Secure AI Alliance and the Open Weights and American AI Leadership initiative, alongside companies including NVIDIA and Microsoft. The Open Secure AI Alliance focuses on developing and sharing open cybersecurity tools and frameworks for AI systems. The Open Secure AI Alliance builds on work already underway through the Linux Foundation's Akrites initiative and OpenSSF, coordinating the remediation and disclosure of vulnerabilities using shared, open tooling. The Open Weights initiative makes the case for openly available models as a driver of innovation and American competitiveness.

Cohesity’s role in this effort is not to build the next open-weight model. It is to help enterprises use open AI safely. That starts with protecting the data those systems depend on and making sure organizations can detect threats and recover when something goes wrong.

This is already part of how Cohesity approaches cyber resilience. REDLab runs live malware in production-grade data protection environments and maps its findings to the MITRE ATT&CK framework. Those findings feed into the threat intelligence that underpins Cohesity’s threat scanning and rapid threat hunting capabilities. Cohesity also draws on intelligence from ecosystem partners, Google Threat Intelligence, CISA, open-source intelligence, customer-supplied indicators, and custom YARA rules to help organizations find malicious artifacts in protected data before that data is restored.

That is Cohesity’s practical contribution to open, secure AI: helping defenders use shared intelligence to protect the enterprise data AI systems retrieve and act on. Open tools and models matter. But enterprises also need confidence that the data they rely on is protected, continually monitored for threats, and recoverable during an attack.

The two efforts address different parts of the AI ecosystem, but they share a premise this piece keeps coming back to: openness does not guarantee security. It has to be built in.

Securing more than the model

Securing the model is only part of the job. Organizations also need to know what data an AI system can access, whether that information is trustworthy, how it is protected, and how quickly it can be restored after an incident. Open defensive tools can help the industry respond faster, but they need to sit on a resilient data foundation.

This is where Cohesity's experience matters. We help organizations protect, secure, govern, and recover the data behind their AI initiatives. In practice, that means protecting training and retrieval data from ransomware and other forms of corruption, controlling access to sensitive information used by agents, and recovering data, applications, and AI environments in the event of an attack. It also means making trusted enterprise data available for AI without weakening the controls around it.

This is the idea behind our Enterprise AI Resilience strategy: organizations need the freedom to move quickly with AI and the resilience to withstand failure, misuse, or attack. The two cannot be separated.

Open models need secure foundation

Enterprises want choice in how and where they run AI. Open-weight models can give them more control over deployment, help keep sensitive data closer to its source, and reduce dependence on a single provider. For organizations with regulatory or data-sovereignty requirements, that flexibility can be especially important.

But open does not automatically mean secure. Running more of the AI stack means an enterprise is responsible for protection, governance, and recovery. Identity controls, data access policies, monitoring, and tested recovery plans do not change because the model is open.

The opportunity lies in combining the flexibility of open AI with the security and operational discipline that enterprises expect from mission-critical systems. That includes the data estate, not just the model.

Through the Open Secure AI Alliance and our support for open-weight AI, Cohesity will advocate for a straightforward principle: AI security starts with the data. It must remain protected from attack, governed according to policy, and recoverable under pressure. That is what makes AI trustworthy in production.