Welcome to The Download, where we break down the day’s biggest tech news into bite-sized, context-filled pieces.

TL;DR: Elon Musk’s AI chatbot Grok now claims its image-generation tools are only available to paying subscribers. The move follows an explosion of nonconsensual sexual imagery on X, which has drawn threats of regulatory action in the UK and EU. But despite the new paywall, there are still easy workarounds that let any X user generate images.

What happened: A December update to Grok’s image-creation feature resulted in a tsunami of sexual and violent images generated by X users. These nonconsensual and explicit images were predominantly of women, but also included children. Seemingly everyone warned that the tool, which is embedded in X, lacked any meaningful guardrails, making it alarmingly easy to create explicit deepfakes at scale.

Governments around the world have raised concerns, and some US lawmakers are pointing to Grok as proof of why stronger protections against deepfake abuse are needed. Yesterday, the European Commission ordered X to preserve documents related to Grok through 2026, as it investigates whether the company is complying with EU law.

Now: Grok has begun telling users who request image generation that it is “currently limited to paying subscribers,” prompting them to sign up. But the restriction appears half-hearted (at best). The Verge found simple workarounds on both desktop and mobile—such as selecting an image and clicking “Edit image.”

What comes next: Grok’s partial retreat shows that even extremely reluctant tech companies will respond when regulators get serious. If scrutiny continues, X could face major fines or even operational limits in key markets. More importantly, this episode could (emphasis on could) accelerate new regulation of AI image-generation tools, especially those involving people’s bodies or likenesses. —SM

20/26 vision

This is where we’ll highlight products, apps, or software that we think you need to know about (or ones to skip). If you have a gadget you love, let us know and we may feature it in a future edition.

It felt like there were more smart glasses at CES this year than there are eyes in the world to wear them. Judging by the popularity of Meta’s Ray-Ban Display glasses—which are selling faster than Meta can make them—it’s clear plenty of companies were eager to create their own copycats. On the CES floor, I tried out two models: the Rokid AI Glasses Style and the Leion Hey2.

Rokid AI Glasses Style

Unlike Meta’s, the Rokid AI Glasses Style don’t actually have a display screen. Instead, everything is handled hands-free via voice commands, activated by: “Hey, Rokid.” The glasses can tap into multiple AI models, including ChatGPT-5 and DeepSeek, to handle tasks like summarizing meetings, giving directions, or identifying an unfamiliar plant you encounter on a hike. The frames include built-in speakers so you can hear the responses, along with a 12-megapixel camera for photos and support for 4K video recording. Design-wise, Rokid is clearly aiming for subtlety. The ones I tried were sunglasses, and they look similar enough to Ray-Bans that a quick glance could fool someone.

The Good: At $299, they’re cheaper than the new Meta Ray-Bans. Rokid says its glasses have up to 12 hours of battery life, while the Meta glasses last up to eight hours for typical use (likely due to their display screen).

The Bad: I don’t like that I need a phone in hand to use these. When I asked it to translate what I was looking at, despite having speakers, it showed me the output on the phone. The AI voice also sounds kind of like the annoying default text-to-voice TikTok narration.

Verdict: Noise

Leion Hey2

Billed as the world’s first AI-powered real-time translation glasses, the Leion Hey2 can translate over 100 languages—Spanish, Arabic, Luxembourgish, you name it. Translations appear almost instantly, with a latency of around 500 milliseconds. Battery life is solid, too: A single charge lasts between six and eight hours, stretching to as much as two weeks with the charging case.

Translations appear on the microLED display in a bright green font reminiscent of The Matrix. Beyond translation, the glasses can also sum up conversations, provide real-time transcriptions, and answer questions via ChatGPT. At $499, they’re much more expensive than the new feature-rich Meta glasses—but those can only do real-time translation in a handful of languages. If you need real-time translation or transcription in as many languages as possible, these might be your best bet.

The Good: The glasses truly are fast. While I didn’t test out super complicated sentences, everything I threw at them was either transcribed or translated quickly (and correctly).

The Bad: Despite the steep price tag, they don't come with a camera or speakers. The text appeared at a pretty far distance, so I’m sure I had a goofy thousand-yard stare while talking to someone right in front of me. They’ve also got more limited capabilities than the new Meta display glasses, which have teleprompter and EMG handwriting features.

Verdict: Signal (if you really need real-time translation in 100+ languages) —WK

When tech and culture intersect, things get interesting. Here, we break down a crossover moment and rate its chaos level.

Chaos Brewing Meter:

Here’s something to add to your to-do list: Get a Chinese peptides dealer. According to a New York Times piece, injecting semi-mystical “performance peptides” is now the latest status hobby in Silicon Valley. (Peptides are like tiny protein pieces that work as simple messages your body uses to tell cells what to do.) People who swear by these jabs use them to allegedly lose weight, focus, sleep better, and heal injuries faster. Some are taking supposedly oxytocin-promoting variants to be better at… making eye contact.

In fairness, some peptides do have clinically proven uses: Insulin and GLP-1s are both peptides. But the Valley spin is to treat them like an IV drip or something you pop in your mouth after a tiring day. There’s currently no FDA approval for using peptides to enter founder flow state—which is why the tech set is sourcing them from Chinese suppliers. Underneath this trend is a familiar Silicon Valley worldview: Regulators are slow, and these mavericks want to move fast and break things. “Things,” in this case, possibly being their own bodies. —WK