Grok's Deepfake Problem and Amazon's Water Confession

Here is perhaps the most uncomfortable timing in recent tech memory: the same week SpaceX is preparing what could be one of the largest IPOs in history, Elon Musk's AI platform Grok is still generating and publicly hosting nonconsensual explicit images of real women.

A review of publicly accessible links on Grok.com turned up dozens of sexualized AI-generated images and videos depicting celebrities and at least one politician. Some are animated. Others are photorealistic enough to pass as real. The subjects never consented to any of it.

This is not a new problem, and that is exactly what makes it so damning. Back in January, Grok made headlines when users on X discovered they could prompt the chatbot to "nudify" photos of women — including, allegedly, images of minors — simply by asking it to redraw someone in a bikini. The backlash was immediate and loud. xAI pledged to roll out safeguards. Lawsuits followed, including a class-action filed in California federal court in March.

Six months later, the safeguards do not appear to be working. Deepfake researcher Henry Ajder, who has spent nearly a decade tracking explicit AI content online, put it plainly: Grok still has not caught up to the basic safety standards that other mainstream AI image tools already have in place.

What makes Grok's situation particularly hard to defend is that this is a solved problem elsewhere. Tools like Midjourney and Adobe Firefly have spent years building and refining content filters specifically designed to block this kind of output. It is not technically impossible. It is a choice about where to invest resources and how seriously to take the harm.

When WIRED reached out to xAI and X for comment, neither responded initially. But shortly after contact was made, the specific images flagged in the investigation disappeared from Grok.com, and the X posts sharing those links were removed for policy violations. That kind of reactive cleanup is notably different from proactive prevention.

xAI did eventually offer a statement insisting it strictly prohibits nonconsensual explicit deepfakes and the use of its tools to undress real people. That policy, however, only holds weight if the platform is actually enforcing it before a journalist calls.

The broader context here matters. The nonconsensual deepfake crisis is not niche. Researchers estimate that the overwhelming majority of deepfake content online is sexual and targets women. Platforms that host generative AI tools carry real responsibility for what those tools produce, especially when the output is this predictable and this preventable.

With SpaceX's IPO putting a spotlight on xAI's parent company, investors will be watching how Musk handles the scrutiny. A company that cannot keep explicit nonconsensual images of real people off its own homepage is a company with a governance problem, not just a moderation one.

Amazon just told the world its data centers drank 2.5 billion gallons of water last year. To put that in perspective, that is roughly enough to fill 3,700 Olympic-sized swimming pools. The company is framing this as good news.

The disclosure came shortly after Seattle passed a one-year moratorium on new data center construction — a move that Amazon's own employees reportedly helped push for. Whether the timing was strategic or just convenient, Amazon is now in the unusual position of releasing water consumption figures it has apparently never made public before, while simultaneously arguing it is more efficient than its rivals.

The numbers Amazon is leaning on do look reasonable on the surface. Its data centers used 0.12 liters of water per kilowatt-hour of electricity in 2025, down two percent from the previous year despite the company expanding its operations significantly. Amazon also published a comparison showing Microsoft, Google, and Meta each using more water per kilowatt-hour than Amazon over recent years.

But that comparison has a catch. The Google figure Amazon is pointing to appears to reflect water usage specifically tied to Gemini AI data centers, while Amazon's number covers its entire global operation. That is not exactly an apples-to-apples fight, and it is the kind of fine print that tends to get buried under a headline claiming industry leadership.

Amazon says about 90 percent of the time its facilities rely on air cooling, only switching to evaporative water cooling during the hottest hours of the hottest days. The company has also been raising the heat tolerance thresholds for its servers, which reduces how often water-intensive cooling kicks in. It claims its data centers are seven times more water-efficient than the industry average, citing a peer-reviewed paper from last year.

Still, there is a significant asterisk on all of this. Amazon's reported figures do not include the water consumed by the power plants supplying electricity to those data centers, nor does it count water used during the construction of new facilities. Indirect water usage from power generation is a substantial and often overlooked part of the full picture, and leaving it out makes the total look considerably cleaner than it might actually be.

The broader context is one that the entire tech industry is grappling with right now. The AI buildout is hungry — for energy, for land, and for water. As cities and states start pushing back with moratoriums and tighter regulations, companies are under growing pressure to justify their footprint with something more than vague sustainability commitments.

Amazon releasing this data at all is a step worth noting. Transparency is genuinely useful, even when the numbers are large. But transparency paired with selective comparisons and incomplete accounting is not quite the same thing as the full picture communities need to make informed decisions about what gets built in their backyards.