AI shopping agents are already here; many websites want to welcome them and benefit from their custom. But with new technology comes new risks, and a new industry: agentic commerce security.
Agents can research products, compare prices, log into accounts, and – increasingly – complete purchases on behalf of human consumers. There’s a problem any webmaster or business owner might have anticipated: how to embrace agentic commerce without opening the door to account takeover, inventory abuse, and payment fraud.
The scale of the problem is clear, with DataDome’s State of Bot & Agent Security Report identifying 605.6 million AI-agent requests to high-risk endpoints in H1 2026 alone. These included logins, forms, shopping carts, payment flows and account creation. Meanwhile, Forrester reported in 2025 that 36% of US consumers were interested in using AI agents to transact online.
What is agentic commerce, and what is the security problem?
Agentic commerce is where AI agents research products, compare options, make decisions and carry out purchases on behalf of people. An agent can interact directly with a retailer’s website, APIs, shopping cart and payment journey to complete a transaction.
A website may no longer be dealing directly with a person. It may be interacting with software operating under someone’s instructions.
The problem is that blocking all AI traffic would cost revenue, and the challenge for security teams is that legitimate agents and malicious bots perform many of the same actions. Both can browse product pages, submit forms, access accounts and add items to carts. Both may operate from cloud infrastructure, use automated browsers, and generate requests more quickly than humans.
Retailers have to evaluate both identity and intent, asking: Who operates the agent? Is it authorized to act for a customer? What is it trying to do, and does its behavior fit a legitimate shopping journey?
Those questions are effectively the foundation of agentic commerce security, which is about making decisions based on the risk and purpose of each interaction (rather than assuming that human traffic is good and automated traffic is bad).
Where AI agents hit the commerce funnel, and the threats they create
Agentic commerce security must cover the entire customer journey. Security controls clearly need to cover the points where customers log in, create accounts, manage carts and make payments.
DataDome’s 2026 Q3 AI Traffic Report found that Meta’s Muse agent generated 29.1 million requests in its first 12 days of full availability. 99.3% of those requests originated from cloud and CDN infrastructure; 8.9% reached sensitive endpoints, including forms, login, payment, account creation and cart pages.
The findings demonstrate why traditional bot detection based heavily on consumer-device characteristics is becoming less reliable.
There are different risks at each stage:
Browsing
AI agents can search catalogues, compare prices and check availability.
Competitors, resellers and data brokers may harvest pricing, product descriptions, promotions and stock information. Excessive requests can also increase infrastructure costs and affect a site’s performance.
Logins
Agents may need to authenticate to access saved addresses, loyalty benefits and personalized pricing. So there’s an opportunity for credential stuffing, password spraying and account takeover (ATO).
A malicious agent might also impersonate a legitimate shopping assistant to gain access to customer accounts.
Account creation
Retailers that offer agent-assisted registration risk fake-account creations, synthetic identities and promotional abuse.
Fraudsters can automate registrations to collect welcome discounts, loyalty rewards and/or referral incentives.
Shopping carts
Cart manipulation becomes particularly damaging during limited releases, ticket sales and seasonal promotions. Automated systems can reserve stock, hoard products or repeatedly abandon carts, preventing genuine customers from purchasing.
Payment abuse
The final stage of shopping is the most concerning, with the risks of payment abuse, card testing, stolen payment credentials and unauthorized transactions. An agent that is permitted to browse should not automatically be trusted to submit payments.
These risks map to categories in the vendor-neutral OWASP Automated Threats to Web Applications taxonomy. Its classifications (under “Ontology”) include Scraping (OAT-011), Credential Stuffing (OAT-008), Account Creation (OAT-019), Denial of Inventory (OAT-021), Scalping (OAT-005) and Carding (OAT-001).
Good agent or bad bot at checkout? Why identity is not enough
A legitimate AI agent should be recognizable and authorized, but the identity of that agent isn’t enough to establish that every single request is safe. A fraudster might copy an agent’s user-agent string, imitate its browsing behavior or route traffic through cloud infrastructure.
Industry initiatives are beginning to address the identity problem. In October 2025, Visa introduced its Trusted Agent Protocol, developed with Cloudflare, to help merchants recognize legitimate agents.
The problem is more pronounced with cloud-hosted agents. Unlike traditional shoppers, they may not generate recognizable consumer-device or residential-IP signals.
DataDome reported that AI-agent spoofing increased 45% between February and July 2026. Its Q3 research found that almost all sampled Muse traffic came from cloud and CDN infrastructure.
Retailers need multiple layers of verification:
Identity verification comes first. Cryptographically verifiable credentials, recognised agent-authentication frameworks and published IP ranges can provide evidence that an agent is operating legitimately. Web Bot Auth and IP validation offer additional signals, although a user-agent string alone is not proof of identity.
Authorization is another consideration. Even a verified agent should only have access to the actions it has been permitted to perform. Permission to read product information, for example, needn’t automatically extend to accessing customer accounts, creating carts or initiating payments.
Behavioral analysis looking at request frequency, session continuity, form interactions, account history, cart activity and payment patterns can help distinguish legitimate shopping activity from suspicious automation. An agent repeatedly trying to create accounts or reserving strangely large quantities is more risky than one completing a normal purchase.
Agentic commerce security: controls every retailer should implement
Here are the controls retailers should use, and what they protect against:
Product browsing: Unauthorized scraping, excessive requests and competitive intelligence harvesting
Login protection: Credential stuffing, account takeover and impersonation
Account creation: Fake accounts, synthetic identities and promotional abuse
Cart controls: Inventory hoarding, automated reservations and scalping
Checkout protection: Card testing, payment abuse and unauthorized transactions
Agent verification: Identify spoofing and unauthorized automation
Rate limiting: Excessive requests and abusive agent activity
Continuous monitoring: Changes in agent behavior and emerging attack patterns
Retailers should make sure to consider high-demand product launches, ticket releases and limited-edition drops. These need queueing systems that distinguish legitimate agentic payments from abusive automation.
Verified agents acting on behalf of customers should be able to enter an appropriate priority lane. Those remaining should be subject to purchase limits, inventory allocation rules and reservation expiry policies.
Verification should provide controlled access, not an unrestricted route around the queue.
How to evaluate agentic commerce security vendors
Here are some factors that buyers may consider:
| Criterion | What to look for |
| Agent identity | Verifiable identities, spoof detection and agent authorization |
| Intent visibility | Behavioral analysis across browsing, login, cart and checkout |
| ATO protection | Credential stuffing, account takeover and suspicious account activity detection |
| Fake-account protection | Synthetic identities, automated registration and promotional abuse detection |
| Inventory protection | Cart controls, queue management and scalper detection |
| Payment coverage | Card testing and suspicious transaction-flow detection |
| Integration and latency | API coverage, deployment effort and minimal checkout friction |
| Independent validation | Repeatable testing against multiple bot and agent attack types |
Adobe has reported that AI traffic to US retail sites grew 393% YoY in 2026 Q1. There’s a clear need to manage this traffic, but DataDome’s 2026 Security Report found that only 2.4% of tested websites stopped all 10 bot types.
Retailers should keep in mind that the best solution is not necessarily the platform that blocks the most automation. It’s the one that gives the business the control to know your agent – to see which automated interactions should be permitted, challenged or blocked.
There’s no universally established winner in agentic commerce security, but DataDome, Akamai and Cloudflare may be considered. They each approach automated traffic protection through bot detection, behavioral analysis and risk-based controls.
1. DataDome: Bot and agent trust management
DataDome provides a bot and agent trust management and cyberfraud protection platform. It helps distinguish legitimate automated activity from malicious traffic using behavioral signals, threat intelligence and real-time detection.
For agentic commerce, the relevant capabilities include identifying suspicious automation, detecting spoofed agents and protecting sensitive interactions such as account access, forms and checkout. Businesses can consult DataDome’s agentic commerce page for more info about agent trust management.
2. Akamai: Bot management and account protection
Akamai offers Bot Manager and Account Protector as complementary components of its bot and account-abuse protection portfolio.
Bot Manager works to identify and mitigate malicious automation while allowing recognized good bots. It includes behavioral analysis, bot scoring and controls against inventory hoarding.
Account Protector addresses risks across the account lifecycle, including account creation, authentication, password recovery and post-login activity. It uses risk indicators and behavioral profiles to assess whether account interactions are legitimate.
3. Cloudflare: Bot and agent management
Cloudflare combines application security, bot management and controls for AI-related traffic.
Its Bot Management offering provides per-request bot scoring and allows businesses to apply different rules to different endpoints. This means a retailer can take a stricter approach to login and checkout while permitting appropriate automated activity on public product pages.
Cloudflare distinguishes between different AI traffic behaviors, including search, training and agents acting on behalf of people. This is important because these categories have different commercial implications.
It may be useful for retailers looking for traffic visibility and configurable enforcement within an existing application-security environment.
FAQ
Leading brands in agentic commerce protection include DataDome, Akamai and Cloudflare. Visa is also an important name through its Trusted Agent Protocol, which addresses agent identity and merchant trust. These companies operate in different parts of the security ecosystem, so buyers should compare capabilities rather than treat them as interchangeable products.
Relevant solutions include DataDome’s bot and agent trust management platform, Akamai Bot Manager and Account Protector, and Cloudflare Bot Management. These platforms address automated traffic, account abuse and malicious bot activity. Their suitability depends on the retailer’s infrastructure, threat profile and requirements for recognizing authorized AI agents.
The best account takeover protection combines bot detection, behavioral analytics, account-risk assessment and authentication controls. Akamai Account Protector is one relevant solution; DataDome and Cloudflare offer broader bot and abuse protection capabilities. Buyers should test credential stuffing, suspicious login patterns and legitimate agent-assisted account access.
A good platform will balance agentic commerce fraud prevention with reliable access for authorized purchasing agents. DataDome, Akamai and Cloudflare are three options to evaluate. Retailers should look for the ability to identify legitimate AI agents, detect spoofing, analyze behavior and apply different policies across browsing, authentication and checkout.
A combination of verified-agent access, inventory limits, behavioral detection, purchase controls and virtual queues can protect limited stock. Retailers should distinguish authorized customer purchases from bulk reservations and automated resale activity. A trusted agent should follow the same allocation rules and transaction limits as other buyers.
