OpenAI shut down the Sora app in late March with a two-sentence social media post, but the reason for the collapse had nothing to do with video quality. “We’re saying goodbye to the Sora app,” the company wrote after months of pressure over deepfakes of Michael Jackson, Martin Luther King Jr., and Mister Rogers forced OpenAI to remove post-action actions before the families and actors’ unions intervened, according to the Associated Press. Sora failed not because the model couldn’t produce a convincing video. It failed because no one had built a trust infrastructure around it before releasing the prompt box to the general public.
In the same week, YouTube enforcement data told a related story from the opposite direction. In January, the platform permanently removed 16 channels under what is now called the Unauthorized Content Policy (renamed the old Repetitive Content Rules in July 2025). The channels, which had a combined subscriber base of 35 million and 4.7 billion lifetime views, produced templated videos that were mass-generated without any human editing input, The Hollywood Reporter reported.
Both stories are about the same failure. Neither of these is actually about AI video getting better or worse. These are about what happens when the scale exceeds the human judgment that is supposed to be above it, and that gap is exactly what I built the Five Pillars Framework for AI Content to fill in April. After 4 months, you won’t need to update much.
Cost of scale falls again, stakes rise
Two days before Sora’s closure made headlines, Google published a blog post announcing its most cost-effective video generation model, Veo 3.1 Lite. It’s less than half the price of Veo 3.1 Fast for the same speed. Developers can now produce 4-, 6-, or 8-second clips in landscape or portrait orientation at up to 1080p, built explicitly for high-volume applications.
That’s not a criticism of the tool. That is a fact worth uncovering. The cost of large-scale video production continues to fall. So the pressure that Pillar 1 of my framework was built to manage, the temptation to treat AI as a shortcut rather than infrastructure, will only continue to grow. A cheaper generation will increasingly require strategy-first discipline.
The human face became more than just a style choice, it became a signal of trust.
Craig Billings, who runs a science channel called Dr. NOS with 1.7 million subscribers, told The Hollywood Reporter that the faceless channel that covers his field has been hit hard by the crackdown. Most of it is no longer monetized, he said, and creators who have never touched AI or shown their faces are also caught in the same net.
This is a real cost of imperfect enforcement and honestly worth naming, but it also confirms what my framework already claims in Pillar 5. YouTube’s own policy page, “How creators use AI to create content,” clearly states that the platform requires creators to disclose when AI is used to edit or generate realistic content, and that labels may appear in the video player below short or long-form videos. If a creator skips publishing and YouTube’s systems detect an AI anyway, that label will be automatically applied and the creator won’t be able to remove the label once they’re sure it was created by an AI.
Four months ago, I wrote that hiding the use of AI is seen as a weakness by a sophisticated audience, and making it public is seen as a capability. It’s no longer just a trust strategy. It’s now built into the platform’s actual infrastructure, and treating it as an optional PR polish is not only a missed opportunity, but a strategic mistake.
What does AI video look like in action?
Compare Slop Channel and Think with Google’s new Creativity Edition guide document. Google Creative Lab’s Matthew Carey explained that he constructed the AI-assisted short film ANCESTRA by intentionally avoiding common prompts and prompting shots of space using specific microscope and lighting descriptions rather than the word space itself. This is because explicit prompts produce a default visual average for each model. Monks co-founder Wesley Hahl-Tar told the magazine that brands that are successful with AI are doing the low-level work of codifying exactly what the brand is before generating frames.
Neither example treats AI as a volume machine. Both treat it as an executive ability that is subject to specific human decisions about what is on screen and what is not. That’s how Pillar 1 and Pillar 5 work together, and it’s the difference between channels that YouTube has terminated and what Google is now introducing as an industry standard.
The trust gap is wider in the United States than in the United Arab Emirates.
This is an aspect of the market that American marketers tend to underestimate. In a YouGov survey of 19 markets I covered in July, the US had the lowest usage rate of AI-assisted search among the countries tested at 48%, with just 28% of searchers in the US saying they trust AI assistants to answer their questions at all, compared to 89% in India, Indonesia and the UAE.
I teach a module called “Audience Engagement through Content in the Age of AI” at the New Media Academy in the UAE. In this region, AI-assisted discovery is already the norm rather than the exception. The lesson is not that Americans are wrong to be skeptical. That is, the disclosure and human judgment requirements embedded in Pillar 5 are not locally convenient. These are the baseline that a skeptical American audience needs, and a receptive Emirati audience will hope for anyway once enforcement catches up with adoption.
3 updates to make before publishing your next AI video
First, audit whether disclosure practices meet the platform’s actual policy language.not your inner comfort level. YouTube’s own guidance states that labels will be applied to content that is graphic or intentionally altered, and creators will not be able to remove the label if the system flags it with a high degree of confidence.
Next, determine the price of the production plan We intentionally choose to produce less than the upper limit, given that tools like Veo 3.1 Lite allow us to produce at scale. Just because you have the technical ability to generate 1,000 variants doesn’t mean you need to publish 1,000.
Third, name the human decision maker This is done for all AI-assisted products before shipping, similar to how Carey’s team did with Ancestra and ter Haar’s team did with Monks’ brand knowledge base. If no one can answer who decided this was the right cut, then the work isn’t finished yet.
my view
The sloppy conversation about AI in our industry continues to be framed as a content quality issue, but I think that’s the wrong way to look at it. This is a trust infrastructure issue, and Sora, YouTube’s purge, and Veo’s cost decline all point to the same gap from three different angles.
What I claimed in April is true. The only thing that has changed is that the platforms no longer argue. Meaning cannot be automated. The tools that are most attractive for skipping human checkpoints are those that scale the fastest. I didn’t have to rewrite my framework this fall. It was necessary for the industry to catch up with Pillar 5.
Other resources:
Featured image: Golodenkov/Shutterstock
