AI tools can provide many potential benefits to e-commerce businesses. They also bring many types of security and privacy risks to the business. AI is pervasive throughout e-commerce sites, generally within five categories of systems. Here are a few of the common privacy and security risks that exist when AI agents are incorporated within each of the indicated categories of systems.  

Payment, financial and fraud management systems. AI agents are increasingly interacting with payment gateways, fraud scoring systems and customer accounts. A few representative risks include model poisoning and adversarial manipulation, over-collection of sensitive financial data, automated false positives, API exploitation and model inversion attacks. 

For example, scammers might trick workers into transferring money using deep fakes of company executives and other staff members during video calls. 

Supply chain and fulfillment systems. AI agents are coordinating inventory, routing, supplier interactions and warehouse robotics. A few representative risks include exposure of supplier and logistics, model-driven operational disruption, IoT and robotics vulnerabilities and third-party data sharing. 

This type of threat can involve threat groups successfully weaponizing open-source AI repositories on platforms, injecting malicious code into popular AI models and extensions. When developers and e-commerce fulfillment systems integrate these models, the poisoned code may exfiltrate sensitive data and credentials. 

Customer service and support systems. AI agents can access personal data, modify accounts, issue refunds or close tickets. A few representative risks include prompt injection and manipulation, unauthorized data exposure, session hijacking, data retention mistakes and hallucinated responses. 

For example, AI might accidentally expose sensitive user data. This could then be used to launch cyberattacks against these users or be sold to others to use for profiling. 

Marketing, personalization and recommendation engines. AI agents are generating content, adjusting pricing, running marketing campaigns, and personalizing experiences. A few representative risks include behavioral profiling and over-personalization, cross-site tracking expansion, data leakage through model outputs, bias and discrimination and unauthorized third-party data sharing.  

Vendor/third-party management and procurement. AI agents are negotiating contracts, placing orders, and evaluating suppliers. A few representative risks include exposure of confidential vendor data, supply chain attacks, harmful automated procurement decisions, integration vulnerabilities, and shadow AI usage by vendors. 

Ensuring that these third-party tools have security methods in place is crucial because even if the primary AI agent is secure, the third-party tools it uses to think or analyze data can become an entry point for attackers and open access to consumer data.  

Here are some security and privacy protections that e-commerce companies can establish, maintain, and enforce in order to provide seamless efficiency and protection to help make AI use beneficial to business, while also protecting against security incidents and privacy breaches:    

Strengthen enterprise authentication. Move employees away from using simple passwords to enforcing multi-factor authentication and single sign-on integrations (where appropriate to the context of use and associated risks) across all procurement and supply chain platforms. Ensure all password management solutions used by the procurement team are enterprise-grade and centrally managed by IT, rather than individual or local solutions. 

Implement privacy protocols. Instruct employees not to share sensitive company information, internal workflow details, or employee or customer data in public B2B forums. Businesses can also implement automated data loss prevention tools that can scan outgoing communications that might expose supply chain vulnerabilities to competitors or bad actors. Ask website visitors to avoid including personal data (theirs or others) in comments and other website locations. 

Persistent data and account monitoring. E-commerce companies should implement automated alerts that monitor for suspicious activity, including potential AI-generated attacks. Along with automated auditing, companies should create a system that emphasizes ongoing, regular audits of account permissions, shipping addresses and authorized buyers. This helps to ensure that company and consumer information is legitimate, accurate and protected.  

Validate vendor domains and portals. Attackers are now using AI-driven tools to create deepfakes or fraudulent written communications to impersonate account managers and redirect wire transfers. Ensure employees have the tools and knowledge to detect AI-generated look-alike vendor websites or emails.  

Install software and patch updates on an ongoing basis. Outdated software and systems are vulnerable to security risks and privacy breaches. Companies should continuously update systems in order to prevent future data breaches and hacks, including those initiated by malicious AI use.   

AI is not only bringing benefits to organizations, but is also creating new types of cybersecurity and privacy risks that have not existed in these forms before. Right now, most e-commerce sites are not implementing comprehensive AI security and controls. In short, e-commerce organizations cannot consider AI agents and other types of tools as just another type of IT product that fits the security and privacy management molds they’ve gotten used to.  

This makes it crucial for companies to be aware of risks, and to take the necessary actions to protect their systems from being compromised and provide protection for consumers.   

Rebecca Herold is the founder, owner, and CEO of Privacy & Security Brainiacs SaaS Services and the Privacy Professor Consultancy, and an IEEE Member.