After AI Reaches Production: 12 Ways Security Teams Can Take Control

In my previous column, I offered some suggestions to help security teams avoid being blindsided when AI applications are moved into production. In this piece, I’d like to offer some thoughts on what is required for security teams to efficiently and effectively incorporate AI applications into the operational security workflow. While there has been much hype around AI applications, many security teams struggle with securing, monitoring, and defending them for a variety of reasons. While not an exhaustive list, I’ve put together 12 practices that I’ve found helpful for incorporating AI applications into the operational security workflow: Visibility: We cannot defend what we cannot see. As such, visibility is really the most fundamental of building blocks when it comes to securing AI applications. Beyond just awareness of and inventorying AI applications, visibility can help us identify exposures of sensitive data, vulnerabilities, deficiencies in controls, fraud, abuse, attacks, and other issues. This makes continuous visibility an extremely important ingredient when it comes to incorporating AI applications into the operational security workflow. Understand Risk: If we’ve taken visibility seriously, we’ll have good data around risk. That data can be used to scientifically understand risk, rather than playing a guessing game. Beyond just a snapshot in time, understanding risk can be done on an ongoing basis in near real-time to allow the security team to more precisely evaluate the risk that one or more applications present to the enterprise. This makes understanding risk another helpful tool when it comes to incorporating AI applications. Build Trust: Discovery is an important part of the visibility piece discussed above. The data generated during discovery can be used to catalyze building relationships between the security team and other important stakeholders, such as application owners, product management, developers, and others. In time, these relationships can mature and a fair amount of trust can be built. This trust will serve the security team very well. Leverage Trust: The trust built in the previous step often facilitates the security team involving itself much earlier in the software development life cycle (SDLC). That makes it much easier to incorporate AI applications into the operational security workflow, which is good news for the overall security posture of the enterprise. Telemetry: If discovery and visibility have been done right, there should be a fair amount of telemetry being generated. It is important that this telemetry data thoroughly cover the AI applications and the infrastructures they’re built upon. This involves generating telemetry from inspecting the AI layer, API layer, and application layer and ensuring that data flow to the SIEM, SOAR, or preferred system of record. Having eyes deep into the application and its infrastructure and making that telemetry available to the security team for analysis, investigation, response, and other steps is extremely important when it comes to properly securing AI applications. Process: While it may be the least sexy one of these points, it is important to develop processes and procedures around securing AI applications, just like we do for all other areas of security. This provides important guidance to the security team and allows them to act and react more agilely – something very important when trying to handle the latest hot potato thrust upon the security team. Enforce: The best controls in the world won’t do an enterprise any good if they can’t enforce those controls. Thus, ensuring that the security team has the ability to easily implement and enforce controls across the various different environments where applications run is an extremely important component when securing AI applications. Preventive Controls: Good preventive controls across the enterprise are going to help security teams with anything, including incorporating AI applications. It is important to remember that these preventive controls should include protection against abuse, fraud, DDoS, malicious automated attacks, and other threats. As such, security teams should ensure they have adequate preventive controls in place, along with the ability to augment and improve those controls in an agile manner. Detective Controls: Continuous Security Monitoring is an important function within any operational security workflow. These detective controls serve as an important partner to preventive controls. However, they require that the requisite visibility is in place and the appropriate telemetry is flowing to the system of record. Assuming this has all been set up properly, it greatly facilitates the security team in its efforts to secure AI applications. Investigation: In the event that a security issue is uncovered, the security team will need to be able to analyze and investigate the corresponding data (logs, events, alerts, etc.). Ensuring that this level of investigation is possible will be another important part of incorporating AI applications into the operational security workflow. It isn’t enough to collect the requisite telemetry – that data need to be available for interrogation by the security team. Mitigation: In the event of a security issue, once the investigation is complete (or at least far enough along), the security team will need to be able to respond, remediate, and recover. This requires that the reach back into the application, API, and AI infrastructure be in place for this to happen. This is something that enterprises often forget about, until they learn the hard way that they can’t mitigate a security issue once they become aware of it. Iterate: Although most of the previous points are more fun than deriving lessons learned, documenting findings, and implementing those lessons learned on a continual basis, these are extremely important. Only through continuous improvement can a security team keep pace with the rapidly changing threat landscape. Enterprises that learn how to iterate successfully will be more nimble when it comes to making the adjustments required to properly secure AI applications as time goes on. AI applications moving into production does complicate things for security teams. That being said, there are steps security teams can take to ease the burden. Hopefully, security teams will be involved earlier in the software development life cycle (SDLC) in the future. Until that time, however, they will need to take steps to proactively prepare to incorporate AI applications into the operational security workflow. Learn More at the AI Risk Summit | Ritz-Carlton, Half Moon Bay WRITTEN BY Joshua Goldfarb Joshua Goldfarb (Twitter: @ananalytical) is currently Field CISO at F5. Previously, Josh served as VP, CTO - Emerging Technologies at FireEye and as Chief Security Officer for nPulse Technologies until its acquisition by FireEye. Prior to joining nPulse, Josh worked as an independent consultant, applying his analytical methodology to help enterprises build and enhance their network traffic analysis, security operations, and incident response capabilities to improve their information security postures. He has consulted and advised numerous clients in both the public and private sectors at strategic and tactical levels. Earlier in his career, Josh served as the Chief of Analysis for the United States Computer Emergency Readiness Team (US-CERT) where he built from the ground up and subsequently ran the network, endpoint, and malware analysis/forensics capabilities for US-CERT. More from Joshua Goldfarb Caught Off Guard: Securing AI After It Hits Production The Hidden ROI of Visibility: Better Decisions, Better Behavior, Better Security The Human IOC: Why Security Professionals Struggle with Social Vetting Security in the Dark: Recognizing the Signs of Hidden Information The Loudest Voices in Security Often Have the Least to Lose The Great Disconnect: Unmasking the ‘Two Separate Conversations’ in Security What Makes a Great Field CXO: Lessons from the Front Lines Perspective: Why Politics in the Workplace is a Cybersecurity Risk Trending Webinar: Third-Party Risk In Practice June 4, 2026 Organizations are investing heavily in third-party risk management, but breaches, delays, and blind spots continue to persist. Join this live webinar as we examine the gap between how organizations think their third-party risk programs are performing and what’s actually happening in practice. Register Virtual Roundtable: CISO Forum 2026 Mid-Year Review June 10, 2026 Explore how attackers are using AI to scale threats and how security teams can respond with AI-driven defenses. Protecting against unmonitored use of generative AI (Shadow AI) in business units and building and enforcing AI governance frameworks. Register People on the Move Opal Security has appointed CPO, CTO, VP of Field Engineering, VP of Marketing, and Head of Product and Solutions Marketing. The Department of the Air Force has appointed Ashley Devoto as Chief Information Officer. Bartley Richardson has been named Chief AI and Autonomous Systems Officer at CrowdStrike. More People On The Move Expert Insights Everybody Is Vibe Coding But Nobody Told The Security Team AI-driven development is not something organizations can or should block. But it must be governed. (Danelle Au) The Zero-Knowledge Threat Actor And The End Of Responsible Disclosure AI can help attackers generate malware, create malicious payloads, bypass simple security checks, and convert vague malicious intent into functional code. (Etay Maor) Raising The Cybersecurity Stakes: Ante Up For The Agentic Era CISOs are now facing machine-speed attacks and asking, “How do I agent?” The industry must provide remediation at scale. (Nadir Izrael) Caught Off Guard: Securing AI After It Hits Production As enterprises rush AI projects into production, security teams are increasingly being forced into reactive mode. (Joshua Goldfarb) Cyber Resilience Is The New Business Continuity Plan The organizations best prepared to face disruption are those that align security, continuity and risk management around what the business cannot afford to lose. (Steve Durbin) Flipboard Reddit Whatsapp Email