Amazon's AI Assistant Can Now Spot Fake Emails

Amazon announced on September 2, 2026 that its Alexa for Shopping assistant can now verify whether a suspicious email, text message, or phone call actually originated from the company — a direct response to a problem that generates 360,000 customer service contacts every year. The feature is live now for U.S. customers, and it represents one of the more concrete anti-scam moves Amazon has made in years.

What Changed

For years, the dominant advice for spotting phishing emails has been a checklist: look for typos, hover over the sender address, check whether the domain is misspelled. That advice is increasingly useless. Scam emails today are well-formatted, grammatically correct, and use sender addresses that differ from the real thing by a single character. Amazon's pitch with this new Alexa for Shopping capability is that it skips the checklist entirely. Instead of asking customers to evaluate the email themselves, it does the comparison internally — against a catalog of billions of messages Amazon has actually sent.

The announcement came with a specific framing from Scott Knapp, Amazon's vice president of worldwide buyer risk prevention: the goal is to put certainty directly in customers' hands, in real time. That word — certainty — matters. Amazon isn't offering a confidence score or a probability. According to both TechCrunch and Amazon's own description of the feature, Alexa for Shopping will only issue a confirmation when it is completely certain the message is genuine. If it can't reach that threshold, it doesn't hedge; it routes the customer to safe defaults, namely checking their orders through the Amazon app or contacting customer support directly.

This binary approach is a meaningful design decision, not just a marketing framing. Most fraud detection systems communicate in probabilities ("this email has a 78% chance of being spam"), which pushes the judgment call back onto the user. By refusing to issue partial verdicts, Amazon is accepting that false negatives — legitimate messages it can't confirm — will exist, in exchange for eliminating false positives that might give scammers a veneer of legitimacy.

Multiple sources confirm the feature covers a broader surface than just email. Phone calls and text messages are also in scope, which matters because Amazon-impersonation scams frequently arrive by SMS and robocall, not just inbox. TechCrunch notes that common scam vectors include fake order confirmations, Prime membership renewal alerts, account suspension warnings, package delivery notifications, and job scams — all categories where the impersonation is convincing enough that Amazon's own customers regularly can't tell.

How It Works

When a customer asks Alexa for Shopping about a suspicious message, the system does more than a simple sender-domain lookup. According to Amazon's own description of the product, the assistant evaluates sender information, message content, formatting, timing, and metadata — essentially everything that can be extracted from a communication — and compares that composite against Amazon's global record of outbound messages. The "catalog of billions" language in Amazon's release is notable: it implies the system isn't just checking a blocklist or running the message through a generic language model. It's doing a lookup against a closed, proprietary corpus that only Amazon can maintain, because only Amazon knows what it actually sent.

That architecture is also what makes the feature genuinely hard to replicate. A third-party email security tool can flag suspicious sender domains and scan for known phishing templates, but it can't tell you whether Amazon sent a specific one-time password to a specific phone number at 3:47 PM last Tuesday. Amazon can, because it has the send logs. Amazon's post describes one example interaction where a customer asks whether a specific OTP text was genuine and Alexa confirms it was — that kind of transaction-level lookup is only possible when the verification system has authoritative access to outbound send records.

Every message submitted for verification is also automatically reported back into Amazon's security intelligence pipeline. Amazon says this will help the system improve over time and supports efforts to identify and take down scam operations. That's a useful side effect: the more customers use the feature, the richer the signal Amazon collects about which fraudulent templates are currently circulating.

What It Means for Developers

The Alexa for Shopping verification feature won't change anything for most developers directly — there's no API announced, no SDK, and no indication Amazon is opening the underlying send-log lookup to third parties. But the design pattern it embeds is worth paying attention to. Certainty-threshold routing — where a system only produces a verdict when it can do so with full confidence, and otherwise fails safe — is a principle that applies anywhere AI is making trust or security decisions. Systems that hedge with probabilistic outputs in high-stakes contexts often make things worse: a "probably legitimate" label can be more dangerous than no label at all, because it gives the user false cover.

Our read is that this is also a quiet admission by Amazon that general-purpose AI assistants are not the right tool for security decisions on their own. Alexa for Shopping can handle this particular verification task precisely because it's not reasoning about the message in isolation — it has privileged access to the ground truth. That's a meaningful constraint. Developers building email security tooling on top of general language models don't have that access, and should be skeptical of any product that claims comparable accuracy without it.

The parallel async path is also worth noting. Earlier this year, Amazon launched [email protected], an email address where anyone — Amazon customer or not — can forward a suspicious message and receive a verification response. No app download, no account login required. TechCrunch and Amazon's own release both describe the two features as complementary rather than redundant: Alexa for Shopping handles real-time, in-conversation verification, while the email address handles cases where someone doesn't have the app or prefers an asynchronous flow. For a problem that affects people across every demographic and technical comfort level, having both paths available is sensible.

The 360,000-contacts-per-year figure Amazon is using as a headline number is worth treating with some skepticism — it counts only people who reached out to customer service, which means the actual number of people confused by potential scams is almost certainly higher. That's the gap this feature is trying to close.

Sources