If you’ve ever stared at the blinking cursor in a chat, trying to figure out the “right” way to tell your boss you’ll be late, or how to comfort a friend without sounding clumsy, WhatsApp now wants to help. This week, the Meta-owned app began rolling out “Writing Help” — an AI feature that can rewrite, polish or change the tone of a draft message for you, with options like professional, funny, supportive or proofread. Tap the new pencil icon while composing in a 1:1 or group chat, pick a style, and the assistant will propose a rewritten version you can edit before sending.

Writing Help is designed to be simple and tightly integrated into the chat flow: you start typing, hit the pencil icon (it appears where the sticker icon used to be for some users), and WhatsApp’s AI generates alternatives you can accept, tweak or ignore. The idea is not to replace your own voice but to rescue you from tone-mismatch moments — the awkward birthday reply, the email-like message that crept into your group chat, the one-liner that needs a kinder landing. The feature is opt-in and appears under a new Private Processing setting in WhatsApp’s Chats menu.

The predictable first question is: can the company read my messages now that they’re sending things to an AI? WhatsApp leans heavily on a privacy engineering stack called Private Processing. The short version: requests go through a privacy-focused cloud path that uses trusted execution environments and cryptographic techniques, so Meta says neither WhatsApp nor Meta can access the plaintext of your messages while the AI works on them — and the company says nothing is stored after the response is generated. Using Writing Help requires enabling Private Processing; if you don’t opt in, the AI features won’t touch your chats.

That claim is similar in spirit to Apple’s Private Cloud Compute — Apple’s own architecture for running heavier AI tasks in the cloud while trying to ensure user data isn’t retained or visible to the company. Both approaches trade off pure on-device processing for more capable cloud models, while building cryptographic and attestation layers intended to keep data confidential. But “it’s private” in principle isn’t the same as “it can’t be abused” in practice, and security experts caution that there are still attack surfaces when data leaves a device, even into a sealed environment.

So what’s different about WhatsApp’s approach?

WhatsApp is not inventing the writing assistant — many apps already offer something similar. Gmail’s “Help me write” and Smart Compose help draft and refine emails; Slack and other messaging platforms have added search, summaries and drafting tools for workplace chat. What WhatsApp brings to the table is twofold: a writing assistant built directly into one of the world’s most used encrypted messenger apps, and an explicit privacy-first deployment path (Private Processing) designed so the company can claim end-to-end protections remain intact even while the cloud does the heavy lifting.

The limits: rollout, languages and who gets it first

For now, Writing Help is rolling out gradually and only in English for users in select markets (the U.S. has been listed among the first). WhatsApp says it hopes to expand the feature to more countries and languages later this year, but there’s no firm timetable. The feature is optional by design and off by default.

Why some people will never use it — and why some already will

The usefulness of a chat-based writing assistant depends on habits and context. For people juggling professional tone inside a messaging app, or for non-native speakers trying to hit a register quickly, a one-tap rephrase is a boon. For others, the overhead of invoking an AI in a rapid conversational back-and-forth may feel clunky — nothing beats the snappy improvisation of a short, human reply. There’s also a social angle: recipients won’t be told a message was AI-assisted, so the tool shifts a small piece of craft from sender to model without explicit disclosure. That’s functionally convenient, but culturally unresolved.

The risks and the trade-offs

Even if Private Processing reduces exposure, experts tell reporters the move trades one set of risks for another. Moving computation to the cloud concentrates processing in fewer physical locations, which may make those targets attractive to attackers; it also requires trust in the correctness of the implementation and in outside attestation and auditing mechanisms. And then there’s the subtle behavioral risk: the more people default to “AI-toned” replies, the more conversational norms could drift toward flatter, more neutral phrasing — or toward AI biases embedded in the models. Wired and other outlets have flagged those concerns in coverage of Private Processing.