Generative AI is already part of everyday work. Employees use it to analyse information, draft content, write code and reduce the time spent on repetitive tasks. In many organisations, however, adoption has moved faster than internal policies and security controls. 

The question is no longer whether employees use generative AI. Security teams need to understand which services are being used, whether they are approved and how organisational policies can be enforced without unnecessarily limiting innovation. 

There is another side to the challenge. Attackers can also use AI to accelerate reconnaissance, vulnerability research, social engineering and exploit development. This leaves organisations with two related but distinct tasks: managing the use of generative AI and protecting systems against AI-assisted attacks. 

Sophos Firewall, Sophos Endpoint and Sophos AI Defense provide complementary controls at the device and network layers to help address both. 

Sophos Firewall: network level AI control 

Every cloud-based AI service and automated workflow relies on network connectivity. This makes the firewall another important point for visibility and policy enforcement. 

Sophos Firewall includes generative AI application signatures within Application Control and its in-line CASB capabilities. These help organisations identify AI services being accessed through the network and apply policies to monitor, block or shape the related traffic.  

The firewall can help security teams determine: 

• which generative AI services are being used; 
• whether users are accessing approved or unsanctioned applications; 
• which users or groups require access; 
• whether a service should be monitored, restricted or blocked; 
• whether AI services may represent a potential data-loss channel. 

Network-level visibility is also valuable for devices that cannot run an endpoint agent, including unmanaged or contractor-owned systems using the corporate network. 

Sophos Endpoint: controlling AI use on devices 

Employees do not access generative AI through a single channel. They may use browser-based services, dedicated desktop applications, coding assistants or AI capabilities embedded in business software. 

Sophos Endpoint provides separate controls for web-based and installed AI applications. 

Controlling AI websites 

Sophos Endpoint Web Control includes a dedicated Generative AI website category. Administrators can allow access, display a warning or block access to services such as ChatGPT and Claude.  

This allows organisations to apply different controls according to their requirements. Approved services can remain available, unfamiliar tools can display a warning and services that do not meet company requirements can be blocked. 

Web Filtering Profiles make these rules easier to maintain across multiple Endpoint and Server policies. Site Lists can also be used to create exceptions or define specific handling for individual AI services.  

Controlling installed AI applications 

Web filtering alone is not enough when employees use dedicated AI clients. 

Sophos Endpoint Application Control includes a GenAI category covering installed tools such as ChatGPT Desktop App, Claude Desktop App, Claude Code, Codex, AnythingLLM and GPT4All. These applications can be detected and, where necessary, blocked from running.  

A sensible starting point is detection rather than immediate blocking. Organisations can first identify the applications in use, assess their business purpose and then decide which tools should be allowed or controlled.  

Protection against AI-assisted attacks 

Endpoint security also plays an important role in defending against attacks created or accelerated with AI. 

AI can help attackers find vulnerabilities and develop or adapt exploits more quickly. However, an exploit still needs to perform recognisable actions on the device, such as executing malicious code, manipulating a trusted process, escalating privileges or stealing credentials. 

Sophos Endpoint is designed to block these exploitation techniques instead of relying only on signatures for individual vulnerabilities. This approach helps protect devices even as the time between vulnerability discovery and potential exploitation becomes shorter. 

Endpoint telemetry also adds valuable context during investigations. It can show which application or process initiated suspicious activity, whether content was downloaded and what actions followed. 

This is increasingly important with AI agents. A user may see an expected result while an AI-driven workflow accesses external content, launches a process or executes a command in the background. Sophos AI Defense and Endpoint telemetry can help connect the AI activity with the resulting device-level behaviour. 

Sophos AI Defense: visibility and governance for AI use 

Sophos AI Defense is designed to help organisations answer a basic but often difficult question: How is AI actually being used across the business? 

The solution provides a central view of AI activity, including the AI applications being accessed, adoption across the organisation, active users and the status of related policies. This helps security teams identify approved and unapproved services and determine where additional controls may be required. 

Visibility is only the first step. Once organisations understand their current AI usage, they can classify services according to business need and risk: 

• Approved: reviewed and accepted for business use. 
• Monitored: visible and logged while the organisation evaluates the service. 
• Restricted: available only to selected users, groups or devices. 
• Blocked: unsuitable because of security, privacy or compliance concerns. 

Sophos AI Defense then works with connected Sophos security controls to turn these decisions into enforceable policies. It integrates with Sophos EDR, XDR and MDR and is available as a separately licensed add-on to these solutions. 

This provides a practical alternative to blocking all generative AI. Organisations can make approved tools available while maintaining visibility into emerging or unsanctioned services. 

Better together 

Sophos AI Defense, Endpoint and Firewall provide different parts of the same picture. 

Sophos Endpoint shows what is happening on the device, including the application and process involved. Sophos Firewall provides the network perspective, showing which services the device contacts and how the traffic is handled. Sophos AI Defense brings AI-related activity and policy information together to provide a clearer view of organisational AI use. 

For example, if an employee installs an unapproved AI client: 

1. Endpoint Application Control can identify the application. 
2. Endpoint telemetry can show when and how it runs. 
3. Firewall Application Control can identify the associated network service. 
4. A relevant policy can restrict or block its use. 
5. Sophos AI Defense can provide central visibility into the activity and policy status. 

This combination helps organisations move from simply discovering AI usage to controlling it in practice. 

A safer approach to AI adoption 

The most effective AI security strategy is not an indiscriminate ban. It is a continuous process: 

1. Discover the AI services and applications already in use. 
2. Assess their business value and potential risks. 
3. Classify them as approved, monitored, restricted or blocked. 
4. Enforce the decision through Endpoint and Firewall controls. 
5. Review usage and policies as technology and business requirements change. 

Sophos AI Defense provides the central AI-focused visibility, Sophos Endpoint delivers device-level control and protection, and Sophos Firewall adds network-level visibility and enforcement. 

Together, they help organisations take advantage of generative AI while establishing the boundaries needed to protect users, systems and business data.