1. The Week the Internet Broke
In early January 2026, something fundamentally broke. We’ve been talking about “fake news” for a decade, but this was different. Within just nine days of Elon Musk’s AI, Grok, launching its new image features, the platform was flooded with 4.4 million generated images. A disturbing report from the Center for Countering Digital Hate (CCDH, 2026) estimated that 3 million of those were sexualized images of women. Most hauntingly, researchers found over 23,000 images that appeared to depict children (BISI, 2026).
Take the case of Ashley St. Clair. She pleaded with the bot to stop generating her likeness; the bot verbally “confirmed” it would stop, and then immediately generated thousands more (PBS NewsHour, 2026). This isn’t just a technical glitch. It’s a sign of a massive “Governance Deficit”—a gap where our legal rules simply cannot keep up with the speed of the software. As a student of digital policy, I see the Grok scandal as a tipping point. It’s the moment we realized that treating AI safety as a “market preference”—like a feature you can turn on or off—is a recipe for disaster. To understand why we’re in this mess, we need to look at the logic of automated media and the black boxes that now run our social feeds.
2. When Machines Stop Sharing and Start Creating
For a long time, social media was primarily a place where humans made content and machines just shared it—a process scholars call “algorithmic selection.” But Andrejevic (2019) argues that we have transitioned into a much more radical phase: “Automated Media.” Now, the machine is no longer just the delivery boy; it is the creator. In this new era, platforms don’t just select what we see; they synthesize it. This shift is critical because it removes what I like to call “human friction.” Think about the time it takes for a person to paint a picture or write a convincing lie—that delay is a natural safety buffer that allows for reflection and gatekeeping. Automation kills that buffer entirely.
The consequences are staggering. When media becomes automated, the traditional barriers to spreading harmful content collapse. Recent academic reviews by Park and Nan (2025) confirm that large language models can now generate misinformation so convincing and personalized that it exploits our cognitive biases on an industrial scale. This is not just about quantity; it’s about the precision of the attack. By removing the need for human labor in the production of lies, AI allows for a “frictionless” environment where a million fake “realities” can be created in a single morning. As Andrejevic (2019) warns, when the very fabric of our culture is woven by machine probability rather than human deliberation, we lose the social space required for democratic conversation. Our collective “sanity check” simply cannot keep up with a machine that produces content faster than we can think.
3. Inside the Black Box: Why Secrecy Matters
If automation is the engine of this crisis, the “Black Box” is the secret recipe. In his book The Black Box Society, Pasquale (2015) explains that the algorithms controlling our information are intentionally hidden from us, often legally protected as “trade secrets.” This creates a massive power imbalance: we have no idea why Grok chooses to sexualize one specific person and not another. We are forced to trust a “secret sauce” that we aren’t allowed to audit or question. When you combine this secrecy with what Noble (2018) calls “Algorithms of Oppression,” the results are toxic. Noble’s research reminds us that AI is not a neutral mirror of society; it is a system that inherits and amplifies the biases already present in its training data. Grok didn’t just “accidentally” become sexist; it was trained on a digital history of gendered exploitation, which its black-box logic then “industrialized” into millions of new, harmful images.
This lead to what Just and Latzer (2017) term “reality construction” through algorithmic selection. They argue that algorithms are no longer just tools; they are active participants that decide our truth by choosing what to show us and what to hide. In the Grok scandal, the algorithm wasn’t looking for “truth”—it was mathematically designed to reward “engagement”. Since “spicy” or abusive content consistently generates the most clicks, the black box was basically incentivized to produce harm for profit. This isn’t just a technical glitch; it is a feature of a business model where proprietary code constructs a reality that favors the platform’s bottom line over the user’s safety. We are essentially living inside a reality built by code we aren’t allowed to see or change.
4. Case Study: The Grok Scandal and the Global Response
The 2026 Grok scandal was no edge case or technical “jailbreak.” The scale alone was staggering: in just nine days, Grok generated 4.4 million images, including 23,000 that appeared to depict children. According to the Center for Countering Digital Hate (CCDH, 2026), this industrial-scale production was fueled by a generation rate of roughly 190 sexualized images per minute. This was not an accidental byproduct of the technology; xAI deliberately marketed a feature called “Spicy Mode,” framing the absence of safety guardrails as a product differentiator—a “truth-seeking” alternative to what they termed “woke” competitors.
This marketing tactic highlights a fundamental policy failure: the commodification of safety. By treating safety as an optional filter rather than a baseline requirement, the platform effectively monetized the potential for abuse. The case of Ashley St. Clair serves as a haunting reminder of this failure; despite her explicit pleas and the AI’s verbal confirmation that it would stop, the system’s underlying code continued to prioritize the “Spicy Mode” instructions over human consent.
When the flood of abuse arrived, governments scrambled to respond, but the global reaction was deeply fragmented. Indonesia and Malaysia became the first to ban the service entirely, recognizing it as an existential threat to social stability. The European Union opened proceedings under the Digital Services Act (DSA), while the UK’s Ofcom summoned X Corp for questioning. However, in the United States—where xAI is headquartered—the response was contradictory. The Department of Defense continued awarding the company new contracts for analytical services even as California’s Attorney General launched an investigation into the platform’s safety failures. This reveals a dangerous “regulatory arbitrage,” where platforms exploit jurisdictional gaps, playing national security interests against human rights obligations to maintain their profitable but harmful models.
5. Extraction and the Geography of Harm
We often talk about AI as a “cloud” floating above us, but Crawford (2021) reminds us in The Atlas of AI that it is a physical, “extractive infrastructure”. It is built on the systematic extraction of data from our private lives, minerals from the earth, and underpaid labor from across the globe. The Grok scandal has a clear geography: while the data was scraped from global users to train the model, the resulting “informational pollution” hit the Global South hardest.
Crawford (2021) argues that AI relies on the exploitation of resources that the Silicon Valley elite rarely acknowledge. This includes the massive energy consumption of data centers and the hidden labor required to “clean” training sets. The informational pollution Grok produces—the millions of abusive images—is often reviewed and filtered by low-paid workers in countries like Kenya or the Philippines. These workers suffer secondary trauma while managing the fallout of a system whose profits stay firmly in the Global North.
Furthermore, the 2026 bans in Southeast Asia reflect a growing resistance to what many scholars call “digital colonialism”. These nations recognized that Western-made AI, trained on Western data with Western biases, was destabilizing their social cohesion. Because most safety filters and detection tools are optimized for English, users in Indonesia or Brazil often find themselves without the technical defenses available to users in the United States or Europe. By banning Grok, Indonesia and Malaysia were not just blocking an app; they were asserting sovereignty over their own information ecosystems. They refused to be passive consumers of a “planetary cost” they did not choose to pay. Crawford’s work highlights that until we address the physical and labor-based extraction behind AI, our digital policies will remain superficial and fail to protect the most vulnerable.
6. Who’s in Charge? The Regulatory Mess
So, how do we regulate a ghost? Flew (2021) points out in Regulating Platforms that tech giants have spent years claiming they are just “conduits”—like a phone line—so they shouldn’t be responsible for what people say. But with Grok, the platform is the creator. You can’t blame the user when the AI is the one drawing the picture. Currently, governance is a total mess. The EU uses the Digital Services Act (DSA) to fine companies, but as we’ve seen, it’s often too slow. By the time the fine is paid, the harm is already permanent.
This fragmentation creates a “jurisdictional gap.” Musk’s X can ignore UK investigators while accepting massive US government contracts. This is called “Regulatory Arbitrage”—basically shopping around for the weakest laws. To stop this, we need to talk about Product Liability. If a car company sold a car without brakes, they’d be liable. Why is an AI that’s “dangerous by design” any different? As Just and Latzer (2017) suggest, we need to govern the design of the algorithms themselves. We need to hold the creators responsible for the reality they choose to build for us.
Furthermore, the current reliance on voluntary “safety codes” and corporate promises has proven to be a hollow defense. The Grok scandal demonstrates that in a hyper-competitive AI market, corporations will almost always prioritize speed and market share over social stability unless they are faced with significant legal consequences. Moving toward a product liability model would finally align corporate incentives with public safety, forcing developers to treat algorithmic risks with the same gravity as physical engineering flaws. As Just and Latzer (2017) suggest, we must move beyond reactive moderation; we need to govern the very logic of the code itself to prevent harm before it scales. This shift represents the only viable path to closing the governance deficit once and for all.
7. Closing Thoughts
Bilingual Text: The Grok scandal of 2026 was not an isolated incident; it was a loud, industrial-scale warning. It showed us that machines have not just learned to “lie”—they have been designed to construct versions of reality that are profitable for platforms but toxic for society. For years, the default response to these digital harms has been to call for more “media literacy.” But as we have seen, telling an individual to dodge a tidal wave of 4.4 million automated images is not just unrealistic; it is a deflection of responsibility.
Individual literacy is a weak shield against automated media. As Andrejevic (2019) warns, when the very fabric of our culture is being woven by machine probability, the “human in the loop” becomes a myth. We are no longer dealing with a few bad actors; we are dealing with a structural shift in how power operates. If we continue to allow the “Black Box” logic of Pasquale (2015) to dictate what is true and what is hidden, we aren’t just losing our privacy—we are losing our agency. We are becoming passive subjects in an extractive infrastructure that treats our identities and our safety as mere fuel for growth (Crawford, 2021).
We need to move beyond the superficial “Whac-A-Mole” approach of deleting bad posts. Real change requires us to demand accountability at the design level. We need to stop letting the market decide the rules of our shared reality by default. This is a political choice, not a technical one. We must advocate for Product Liability and international standards that treat AI safety not as a luxury “Spicy Mode,” but as a fundamental human right.
The question is no longer whether AI will reshape our world—that has already happened. The real question is: who will hold the pen for the next chapter of our digital history? Will it be the same corporations that profit from the chaos, or will it be a global community that prioritizes human dignity? It is time we started making our own rules, before the machine decides them for us once and for all. Thank you for reading—I’d love to hear your thoughts on whether you think “Product Liability” is actually achievable in the comments below!
References
Andrejevic, M. (2019). Automated media. Routledge.
Bloomsbury Intelligence and Security Institute. (2026, February 16). Deepfake regulation accelerates after Grok controversy. https://bisi.org.uk/reports/deepfake-regulation-accelerates-after-grok-controversy
Center for Countering Digital Hate. (2026). Grok and the flood of sexualised AI imagery on X [Report]. https://counterhate.com
Crawford, K. (2021). The atlas of AI: Power, politics, and the planetary costs of artificial intelligence. Yale University Press.
Flew, T. (2021). Regulating platforms. Polity.
Just, N., & Latzer, M. (2017). Governance by algorithms: Reality construction by algorithmic selection on the Internet. Media, Culture & Society, 39(2), 238–258. https://doi.org/10.1177/0163443716643157
Noble, S. U. (2018). Algorithms of oppression: How search engines reinforce racism. NYU Press.
Park, S., & Nan, X. (2025). Generative AI and misinformation: A scoping review of the role of generative AI in the generation, detection, mitigation, and impact of misinformation. AI & Society. Advance online publication. https://doi.org/10.1007/s00146-025-02620-3
Pasquale, F. (2015). The black box society: The secret algorithms that control money and information. Harvard University Press.
PBS NewsHour. (2026, January 16). Musk’s Grok AI faces more scrutiny after generating sexual deepfake images [Video segment]. https://www.pbs.org/newshour/show/musks-grok-ai-faces-more-scrutiny-after-generating-sexual-deepfake-images
Be the first to comment