Pangram Brings AI Detection Into Your Inbox

Pangram, a company that develops artificial intelligence text-detection tools, announced on Wednesday, September 16, a new Gmail integration that automatically checks incoming messages and classifies them as "AI," "Mixed," or "Human." The tool, already known for analyzing documents and web pages, now enters one of the most contested frontiers of the generative AI era: the email inbox.

The integration requires the Individual plan, starting at US$ 20 per month, and functions as part of Pangram's ecosystem of products — which includes a text detector, image detector, and similarity checker. After connecting your account via Google's OAuth, the extension scans your 100 most recent conversations and then continues checking messages as they arrive. The result appears as a colored label next to the email, visible without opening the message.

How it works in practice

Pangram claims its detector operates with a false positive rate of one in ten thousand and that an independent study conducted by the Becker Friedman Institute at the University of Chicago, in collaboration with researchers from the University of Maryland, classified the tool as the most accurate on the market. According to data disclosed by the company on its website, the detector analyzed 244 conversations in one week, labeling 163 (67%) as human, 24 (10%) as mixed, and 57 (23%) as AI-generated.

The logic behind these numbers seems straightforward: the more time people spend on emails, the more likely they are being written by machines. Pangram offers three action options after the label is applied: keep the email in the inbox, automatically archive it, or send it to the spam folder. Users can also exclude specific domains, labels, or senders from scanning.

The approach has a practical utility. Max Spero, Pangram's co-founder and CEO, told Gizmodo he has been using the tool for about two months and that it has been "pretty huge to be able to see cold emails labeled as AI." His personal preference is to label AI-generated emails and leave them in the inbox, but he mentioned this could change "if AI emails ramp up."

The challenge of authorship detection

But detecting AI-generated text is not as straightforward as it might seem. Pangram cannot distinguish between an email written entirely by an AI and one written by a human who used AI to refine clarity or tone. This raises an important question: what is the difference between using AI to "help" write a professional email and using AI to generate spam?

Keyvan, editor of the blog Keyvan's blog, argued in an October 2026 analysis that labeling emails as "AI" is equivalent to labeling them as "spam," when they are actually different things. "Real people use AI to write emails. Often because they're short on time and worried that a hurried email will come across as curt, or because they want to improve clarity, or because they're not writing in their native language," he wrote, testing the tool with legitimate emails that were correctly marked as AI — but that clearly were not spam.

This ambiguity is the weak point of any authorship detection tool. Pangram itself acknowledges that "we'll never get to one hundred percent accuracy" according to Spero — which the company describes as "accumulating more certainty using more data." But until the technology evolves to distinguish intent from authorship, the "AI" label will remain partial information that the user needs to interpret.

Privacy and the cost of sending data to third parties

Another concerning aspect is the privacy model. Although Pangram claims it does not store email contents after analysis and does not use these data to train its models, the user is still sending the full content of all received messages to a third-party server. The company says the detector "does not receive information about the sender, recipient, subject, or other mailbox metadata," but the body of the email itself is sent for classification.

This creates a paradox: the tool that promises to protect the user against unwanted emails requires them to trust an external company with their private mail. Pangram offers full control — the user can disconnect the integration at any time — but trust in a private company is different from the built-in protection offered by Gmail itself.

What this means for the future of email

What Pangram demonstrates is that AI detection is moving from the domain of academic tools and editorial content into professional daily life. As generative AI becomes a standard part of billions of people's workflow — whether for writing emails, responding to messages, or creating documents — the question of how to identify machine-generated text will cease to be a curiosity and become a practical necessity.

The AI email assistant market is already fragmented into several categories: Gemini within Gmail, Shortwave at $24/month, SaneBox at $15/month, Superhuman Business at $15/member. Pangram enters a different category — not as a composition assistant, but as an authorship filter. It is a different approach that reflects the recognition that, at some point, users will need to know not just what to write, but who (or what) wrote it.

Sources: WIRED, Gizmodo, ATLABYTE

✓ Independent sources cross-checked and verified before publishing