Casey Newton identifies a startling inversion in the digital landscape: the tools that create synthetic media are becoming more restrictive, while the platforms that distribute them are opening their doors. This shift suggests that the primary battleground for deepfake regulation has moved from the point of creation to the point of publication, leaving victims to police their own reputations.
The Great Inversion
Newton begins by dissecting the historical model of content moderation, where the burden fell heavily on platforms to remove harmful content after it was posted. He notes that in the first era of digital media, "US users carry the legal liability for most of what they post, thanks to Section 230," while platforms acted largely to protect their business interests rather than out of moral obligation. The arrival of generative artificial intelligence, however, introduced a new actor: the tool itself. Newton observes that major AI developers like OpenAI and Anthropic have adopted a cautious stance, restricting their tools from generating political messages or sexual imagery even if that content never leaves the user's device. As Newton puts it, "In the old days, digital tools were permissive, and digital platforms were restrictive. Today, digital tools are restrictive."
This framing is crucial because it highlights a disconnect in the safety architecture. While the creation of a deepfake might be blocked at the source by a chatbot or image generator, the distribution mechanisms remain wide open. Newton argues that this creates a dangerous asymmetry where the most accessible tools for creating synthetic media are more restrictive than the tools for distributing it. Critics might note that this analysis assumes a level of enforcement consistency among AI companies that may not exist in practice, as users often find workarounds or use open-source models that lack these guardrails. Nevertheless, the observation that the "leverage in moderating deepfakes may not lie where we expected" is a compelling starting point for understanding the current regulatory vacuum.
In the old days, digital tools were permissive, and digital platforms were restrictive. Today, digital tools are restrictive.
YouTube's Green Light
The commentary then turns to YouTube's recent policy announcement, which Newton characterizes not as a set of prohibitions but as a "general green light for people to post synthetic media widely on YouTube." The platform's approach relies on labeling rather than removal, requiring metadata tags for synthetic content and overlay labels for sensitive topics. Newton points out the significant gap in enforcement: "For the most part, though, you'll be able to post synthetic media on YouTube as you wish."
This policy effectively shifts the burden of harm mitigation onto the individuals depicted in the videos. Newton writes, "The responsibility for addressing harm here falls not on the platform, or the user, or the tool, but on the victim." Under this framework, a person whose likeness is used without consent must fill out a form and hope for a favorable review, with YouTube promising to "consider a variety of factors" without defining what those factors are. This vagueness is a major vulnerability in the policy. While Newton acknowledges that the policy carves out meaningful permissions for satire and parody, he warns that "the better that synthetic media gets, the weaker this policy feels."
The distinction Newton draws between the treatment of major-label musicians and average citizens is particularly sharp. He notes that major-label artists will receive special protections, able to request takedowns of voice-mimicking videos even in parodies, a move he describes as a "gift designed to buy some goodwill" from record labels. This creates a two-tiered system of protection where fame and corporate backing determine one's safety from deepfakes. A counterargument worth considering is that over-regulating satire could stifle legitimate political discourse, but Newton's concern that this creates a "new battlefield" for harassment feels well-founded given the platform's history with inconsistent enforcement.
The Distribution Problem
Newton concludes by situating YouTube's move within the broader ecosystem, noting that Meta's synthetic media policy is similarly permissive, while TikTok stands alone in banning synthetic videos of non-public figures. The core of his argument is that the most likely source of harm will not come from the leading AI companies that block content at the source, but from "open-source tools used to create materials hosted and distributed by tech giants." He warns that if social networks hope to avoid a repeat of the post-2016 tech backlash, they must proceed carefully. The piece ends on a note of uncertainty, describing the policy as a "first draft" that will surely evolve as threat models are constructed.
The most accessible tools we have for creating synthetic media are more restrictive than the tools we have for distributing it.
Bottom Line
Newton's analysis effectively exposes a critical flaw in the current approach to AI governance: by focusing on restricting creation, the industry has inadvertently empowered distribution channels to become the primary vectors for harm. The strongest part of this argument is the clear identification of the burden-shifting onto victims, which is a precarious and likely unsustainable strategy. The biggest vulnerability lies in the assumption that labeling will suffice to mitigate the chaos of realistic synthetic media, a hope that may not hold as the technology becomes indistinguishable from reality.