Be part of our day by day and weekly newsletters for the newest updates and unique content material on industry-leading AI protection. Be taught Extra
Balancing the paradox of defending one of many world’s main journey, software program and providers companies towards the accelerating threats of AI illustrates why CISOs have to be steps forward of the newest adversarial AI tradecraft and assault methods.
As a number one international B2B journey platform, American Categorical International Enterprise Journey (Amex GBT) and its safety workforce are doing simply that, proactively confronting this problem with a twin concentrate on cybersecurity innovation and governance. With deep roots in a financial institution holding firm, Amex GBT upholds the very best knowledge privateness requirements, safety compliance and threat administration. This makes safe, scalable AI adoption a mission-critical precedence.
Amex GBT Chief Data Safety Officer David Levin is main this effort. He’s constructing a cross-functional AI governance framework, embedding safety into each part of AI deployment and managing the rise of shadow AI with out stifling innovation. His strategy gives a blueprint for organizations navigating the high-stakes intersection of AI development and cyber protection.
The next are excerpts from Levin’s interview with VentureBeat:
VentureBeat: How is Amex GBT utilizing AI to modernize menace detection and SOC operations?
David Levin: We’re integrating AI throughout our menace detection and response workflows. On the detection aspect, we use machine studying (ML) fashions in our SIEM and EDR instruments to identify malicious conduct quicker and with fewer false positives. That alone accelerates how we examine alerts. Within the SOC, AI-powered automation enriches alerts with contextual knowledge the second they seem. Analysts open a ticket and already see crucial particulars; there’s not a must pivot between a number of instruments for primary info.
AI additionally helps prioritize which alerts are doubtless pressing. Our analysts then spend their time on the highest-risk points moderately than sifting via noise. It’s a large increase in effectivity. We will reply at machine velocity the place it is sensible, and let our expert safety engineers concentrate on advanced incidents. In the end, AI helps us detect threats extra precisely and reply quicker.
VentureBeat: You additionally work with managed safety companions like CrowdStrike OverWatch. How does AI function a drive multiplier for each in-house and exterior SOC groups?
Levin: AI amplifies our capabilities in two methods. First, CrowdStrike OverWatch offers us 24/7 menace searching augmented by superior machine studying. They continually scan the environment for delicate indicators of an assault, together with issues we’d miss if we relied on handbook inspection alone. Meaning we now have a top-tier menace intelligence workforce on name, utilizing AI to filter out low-risk occasions and spotlight actual threats.
Second, AI boosts the effectivity of our inside SOC analysts. We used to manually triage way more alerts. Now, an AI engine handles that preliminary filtering. It will probably shortly distinguish suspicious from benign, so analysts solely see the occasions that want human judgment. It appears like including a wise digital teammate. Our workers can deal with extra incidents, concentrate on menace searching, and choose up superior investigations. That synergy—human experience plus AI assist—drives higher outcomes than both alone
VentureBeat: You’re heading up an AI governance framework at GBT, primarily based on NIST rules. What does that appear like, and the way do you implement it cross-functionally?
Levin: We leaned on the NIST AI Danger Administration Framework, which helps us systematically assess and mitigate AI-related dangers round safety, privateness, bias and extra. We shaped a cross-functional governance committee with representatives from safety, authorized, privateness, compliance, HR and IT. That workforce coordinates AI insurance policies and ensures new tasks meet our requirements earlier than going dwell.
Our framework covers your entire AI lifecycle. Early on, every use case is mapped towards potential dangers—like mannequin drift or knowledge publicity—and we outline controls to deal with them. We measure efficiency via testing and adversarial simulations to make sure the AI isn’t simply fooled. We additionally insist on at the very least some degree of explainability. If an AI flags an incident, we wish to know why. Then, as soon as programs are in manufacturing, we monitor them to substantiate they nonetheless meet our safety and compliance necessities. By integrating these steps into our broader threat program, AI turns into a part of our total governance moderately than an afterthought.
VentureBeat: How do you deal with shadow AI and guarantee staff observe these insurance policies?
Levin: Shadow AI emerged the second public generative AI instruments took off. Our strategy begins with clear insurance policies: Staff should not feed confidential or delicate knowledge into exterior AI providers with out approval. We define acceptable use, potential dangers, and the method for vetting new instruments.
On the technical aspect, we block unapproved AI platforms at our community edge and use knowledge loss prevention (DLP) instruments to forestall delicate content material from being uploaded. If somebody tries utilizing an unauthorized AI website, they get alerted and directed to an accepted different. We additionally rely closely on coaching. We share real-world cautionary tales—like feeding a proprietary doc right into a random chatbot. That tends to stay with folks. By combining person training, coverage readability and automatic checks, we will curb most rogue AI utilization whereas nonetheless encouraging reputable innovation.
VentureBeat: In deploying AI for safety, what technical challenges do you encounter, for instance, knowledge safety, mannequin drift, or adversarial testing?
Levin: Knowledge safety is a major concern. Our AI typically wants system logs and person knowledge to identify threats, so we encrypt these feeds and limit who can entry them. We additionally be sure no private or delicate info is used except it’s strictly essential.
Mannequin drift is one other problem. Assault patterns evolve continually. If we depend on a mannequin educated on final 12 months’s knowledge, we threat lacking new threats. Now we have a schedule to retrain fashions when detection charges drop or false positives spike.
We additionally do adversarial testing, basically red-teaming the AI to see if attackers might trick or bypass it. Which may imply feeding the mannequin artificial knowledge that masks actual intrusions, or making an attempt to govern logs. If we discover a vulnerability, we retrain the mannequin or add further checks. We’re additionally massive on explainability: if AI recommends isolating a machine, we wish to know which conduct triggered that call. That transparency fosters belief within the AI’s output and helps analysts validate it.
VentureBeat: Is AI altering the position of the CISO, making you extra of a strategic enterprise enabler than purely a compliance gatekeeper?
Levin: Completely. AI is a first-rate instance of how safety leaders can information innovation moderately than block it. As an alternative of simply saying, “No, that’s too dangerous,” we’re shaping how we undertake AI from the bottom up by defining acceptable use, coaching knowledge requirements, and monitoring for abuse. As CISO, I’m working carefully with executives and product groups so we will deploy AI options that really profit the enterprise, whether or not by enhancing the shopper expertise or detecting fraud quicker, whereas nonetheless assembly rules and defending knowledge.
We even have a seat on the desk for giant selections. If a division needs to roll out a brand new AI chatbot for journey reserving, they contain safety early to deal with threat and compliance. So we’re transferring past the compliance gatekeeper picture, moving into a job that drives accountable innovation.
VentureBeat: How is AI adoption structured globally throughout GBT, and the way do you embed safety into that course of?
Levin: We took a worldwide heart of excellence strategy. There’s a core AI technique workforce that units overarching requirements and tips, then regional leads drive initiatives tailor-made to their markets. As a result of we function worldwide, we coordinate on finest practices: if the Europe workforce develops a sturdy course of for AI knowledge masking to adjust to GDPR, we share that with the U.S. or Asia groups.
Safety is embedded from day one via “safe by design.” Any AI mission, wherever it’s initiated, faces the identical threat assessments and compliance checks earlier than launch. We do menace modeling to see how the AI might fail or be misused. We implement the identical encryption and entry controls globally, but in addition adapt to native privateness guidelines. This ensures that regardless of the place an AI system is constructed, it meets constant safety and belief requirements.
VentureBeat: You’ve been piloting instruments like CrowdStrike’s Charlotte AI for alert triage. How are AI co-pilots serving to with incident response and analyst coaching?
Levin: With Charlotte AI we’re offloading a variety of alert triage. The system immediately analyzes new detections, estimates severity and suggests subsequent steps. That alone saves our tier-1 analysts hours each week. They open a ticket and see a concise abstract as a substitute of uncooked logs.
We will additionally work together with Charlotte, asking follow-up questions, together with, “Is that this IP deal with linked to prior threats?” This “conversational AI” side is a significant assist to junior analysts, who study from the AI’s reasoning. It’s not a black field; it shares context on why it’s flagging one thing as malicious. The online result’s quicker incident response and a built-in mentorship layer for our workforce. We do keep human oversight, particularly for high-impact actions, however these co-pilots allow us to reply at machine velocity whereas preserving analyst judgment.
VentureBeat: What do advances in AI imply for cybersecurity distributors and managed safety service suppliers (MSSPs)?
Levin: AI is elevating the bar for safety options. We count on MDR suppliers to automate extra of their front-end triage so human analysts can concentrate on the hardest issues. If a vendor can’t present significant AI-driven detection or real-time response, they’ll wrestle to face out. Many are embedding AI assistants like Charlotte immediately into their platforms, accelerating how shortly they spot and include threats.
That mentioned, AI’s ubiquity additionally means we have to see previous the buzzwords. We check and validate a vendor’s AI claims—“Present us how your mannequin discovered from our knowledge,” or “Show it might deal with these superior threats.” The arms race between attackers and defenders will solely intensify, and safety distributors that grasp AI will thrive. I totally count on new providers—like AI-based coverage enforcement or deeper forensics—rising from this development.
VentureBeat: Lastly, what recommendation would you give CISOs beginning their AI journey, balancing compliance wants with enterprise innovation?
Levin: First, construct a governance framework early, with clear insurance policies and threat evaluation standards. AI is just too highly effective to deploy haphazardly. When you outline what accountable AI is in your group from the outset, you’ll keep away from chasing compliance retroactively.
Second, accomplice with authorized and compliance groups upfront. AI can cross boundaries in knowledge privateness, mental property, and extra. Having them onboard early prevents nasty surprises later.
Third, begin small however present ROI. Decide a high-volume safety ache level (like alert triage) the place AI can shine. That fast win builds credibility and confidence to broaden AI efforts. In the meantime, put money into knowledge hygiene—clear knowledge is all the pieces to AI efficiency.
Fourth, practice your folks. Present analysts how AI helps them, moderately than replaces them. Clarify the way it works, the place it’s dependable and the place human oversight continues to be required. A well-informed workers is extra prone to embrace these instruments.
Lastly, embrace a continuous-improvement mindset. Threats evolve; so should your AI. Retrain fashions, run adversarial assessments, collect suggestions from analysts. The expertise is dynamic, and also you’ll must adapt. When you do all this—clear governance, robust partnerships, ongoing measurement—AI might be an unlimited enabler for safety, letting you progress quicker and extra confidently in a menace panorama that grows by the day.
VentureBeat: The place do you see AI in cybersecurity going over the following few years, each for GBT and the broader {industry}?
Levin: We’re heading towards autonomous SOC workflows, the place AI handles extra of the alert triage and preliminary response. People oversee advanced incidents, however routine duties get totally automated. We’ll additionally see predictive safety—AI fashions that forecast which programs are most in danger, so groups can patch or phase them upfront.
On a broader scale, CISOs will oversee digital belief, guaranteeing AI is clear, compliant with rising legal guidelines and never simply manipulated. Distributors will refine AI to deal with all the pieces from superior forensics to coverage tuning. Attackers, in the meantime, will weaponize AI to craft stealthier phishing campaigns or develop polymorphic malware. That arms race makes strong governance and steady enchancment crucial.
At GBT, I count on AI to permeate past the SOC into areas like fraud prevention in journey bookings, person conduct analytics and even customized safety coaching. In the end, safety leaders who leverage AI thoughtfully will achieve a aggressive edge—defending their enterprises at scale whereas releasing expertise to concentrate on probably the most advanced challenges. It’s a significant paradigm shift, however one which guarantees stronger defenses and quicker innovation if we handle it responsibly.
