Could Google’s Project Mariner Totally Change the Way We Manage Fraud? Here’s What We Know.
Google just raised the stakes in the automated assistant game.
Last month, the tech giant revealed the latest version of Project Mariner, a still-experimental AI agent that can browse the web and take actions on the user’s behalf. It wasn’t exactly a surprise; the project has been in development for years (and the launch date pushed back for months). Still, this working prototype offers significant advances compared to what developers saw last year.
Project Mariner has the potential to upset and radically reshape eCommerce. So, I wanted to take a moment to explore the technology, its uses and its challenges, and its potential impact on fraud detection and chargeback management.
Recommended reading
- What is Alipay? Can US Consumers & Merchants Use Alipay?
- Dispute Apple Pay Transaction: How Does The Process Work?
- Terminal ID Number (TID): What is it? What Does it Do?
- What are NFC Payments? Security, Implementation & More
- How Integrated Payments Work: Benefits, Use Cases, & More
- How EMV Chip Cards Work: Pros, Cons, Data Points & More
What is Project Mariner?
Project Mariner is a Gemini-powered extension that can browse, navigate, and use the web in a way that mimics human behavior. Mariner can move the cursor around the screen, click buttons, scroll, and fill out forms as needed. It also has the capability to perform research, make bookings or reservations, and a lot more.
It’s like having your own digital gofer. Think Google Assistant, cranked up to 11.
While Project Mariner isn’t commercially available yet, it has been rolled out to an expanded group of developers, as well as Google AI Ultra subscribers in the US. Experts are calling it a UX paradigm shift, where a generative AI system interacts with websites more deeply than the actual users.
The original version of Google’s service was clunky. It was limited to handling one task at a time, and only on the frontmost tab. This latest iteration runs on virtual machines in the cloud through what Google calls a “system of agents.” Translated, that means Mariner can now do its thing in the background while still allowing the user to work on something totally unrelated. According to Google, the new and improved Project Mariner can handle as many as ten tasks simultaneously.
How Does Project Mariner Work?
Project Mariner goes by what’s displayed in the user’s Chrome browser. It can, of course, read text, but it can also identify and understand code, images, buttons, forms, and more. This gives the AI agent an overall understanding of the entire page; not just the text.
The user can make requests in natural language and have Mariner handle otherwise work-intensive tasks, like automating data entry, streamlining repetitive actions in onboarding, providing customer support, and even researching competitors.
On the consumer side, Mariner can find recipes, create shopping lists, remember and apply preferences, and so on. More important to retailers, the AI agents can also help consumers compare prices and reviews, automate more of the checkout process (including applying loyalty discounts or coupons), and provide real-time updates on fulfillment, shipping, and delivery. That last item alone could have a huge impact on fraud management.
Project Mariner: Implications for Fraud Management
In the interest of security, Google has put some limitations on what Mariner can do. It’s not allowed to fill out credit card numbers or provide billing information, for example, so it can’t make a purchase without user input. The agent also can’t authorize things like a terms of service agreement, nor will it accept cookies on the user’s behalf.
That’s all good. But, we’re talking about things that make the system safe to use as intended. That’s not necessarily going to stop fraudsters from finding ways of twisting Mariner’s AI to their own ends.
Crooks are already subverting services like Google Assist to scam users by bypassing voice recognition, facial verification, and visual liveness checks. The technology for creating authentic looking deepfakes has been shown to potentially fool even the most discerning viewers.
It stands to reason that expanding the capabilities of AI in digital assistants may open up even more opportunities for fraudsters. Malicious actors could potentially make their spoofed scams even more realistic, or extend their reach.
And, for all its accuracy, even advanced machine learning must have good data to work with. Mariner is still subject to hallucinations, which could be driven by fraudsters, or used to trick the AI into believing a bogus purchase is authorized.
While it might seem counterintuitive, the privacy protocols built into Mariner could potentially work against it. Transactions may end up being more opaque, which could protect data but would almost certainly hamper fraud detection.
Advanced AI assistants could be a boon for consumers, but it could cause new headaches for merchants.
Let’s talk about how we could help.
Request a Demo
Could Mariner Impact Chargeback Liability & Response?
Who should be responsible for chargeback losses stemming from new AI technology? If you asked any of the parties involved, you’d probably get the same answer: “Not me!”
Banks and merchants already feel like they bear too much of the burden. Card networks have the power to stay out of the argument, and most consumers fail to even see a problem. That leaves all fingers pointing at Google. But, it’s doubtful that all of the liability would land there, or if the tech giant would be blamed at all.
So… who gets the bill?
For most valid claims, the liability would likely stay wherever it is now. Liability in cases of true criminal fraud, for example, would probably still lie with the least-secure party in the transaction. But, services like Mariner could put us in all new territory. For instance, what happens if an AI agent makes a purchase that later gets disputed?
Google says that can’t happen, that the user would have to OK the purchase. It’s hard to believe people won’t find ways to circumvent that. The fact that Google built in protections against it would probably shift liability away from them, though.
My best guess is that such transactions would be treated like family fraud, which means the merchant will be required to prove that the claim is unjustified. AI has the potential of impacting that, as well; it could help fraudsters create deepfaked evidence that would be harder to contest.
Opportunities for Forward-Thinking Merchants
In terms of fraud and chargebacks, the impact of services like Project Mariner certainly won’t be all bad news. Yes, AI fraud is quickly evolving, but so is AI detection and prevention technology.
Advanced tools now program multi-angle face scans and 3D depth-sensing into their anti-spoofing algorithms. Deepfake detection AI analyzes visual inconsistencies and can detect synthetic overtones across the entire audio spectrum. The technology can help after a transaction, too; in cases of first-party fraud, AI can be used to automatically gather compelling evidence, create a case, and even re-present transactions.
The biggest issue right now is that we don’t know what the full impact of Mariner will be. We won’t know until it — and similar services — really get out into the wild.
But, we do know that it’s coming, and “forewarned is forearmed,” as they say. The time for merchants to adapt is now; rethinking not only their defenses, and how they need to communicate with a new kind of “customer.”
Project Mariner Is a Wake-Up Call
Project Mariner isn’t just a cool tech demo; it’s a signal that AI is no longer on the sidelines of commerce.
Merchants need to start reimagining their fraud and chargeback management strategies now. Traditional rule-based prevention models will prove to be no match for sophisticated AI-powered fraud. Simply learning to differentiate between genuine AI requests and fraudulent attempts could stump current technology.
More complex systems like Project Mariner will logically lead to more complex fraud techniques. That, in turn, will require increasingly complex fraud prevention solutions with the capability to learn and adapt to a shifting fraud landscape.
AI is already becoming an essential tool in the fight against fraud. But, organizations will ultimately need to embrace a more comprehensive, multifaceted approach.
