OpenAI's Sora arrived in November 2025 with considerable momentum. The invitation-only launch strategy and the technology's capacity to generate strikingly lifelike videos created immediate demand. Users could upload their faces and voices, enabling others to create deepfakes featuring them in virtually any scenario. Yet this week's announcement that OpenAI would shut down the application merely six months into operation signals the end of what many had viewed as a potential challenger to established social platforms.

For researchers tracking the proliferation of AI-generated content, the closure brought relief. Sora functioned as a consolidated platform merging social media's engagement mechanics with generative AI's capacity to produce persuasive fabricated videos. The combination proved potent: within months, the service became a major source of warzone disinformation and an avalanche of synthetic content that inundated researchers, journalists, fact-checkers, law enforcement, and ordinary people with verification requests.

Yet Sora's failure carries broader implications for the technology sector and its trillion-dollar AI enterprises. Understanding what went wrong illuminates critical vulnerabilities in the current approach to artificial intelligence deployment.

Why Sora failed

OpenAI's wager that combining artificial intelligence with social media dynamics would generate substantial profits proved miscalculated. Several structural problems emerged:

  1. Conventional social platforms shift content creation costs and labor onto users; Sora bore all expenses for video generation internally. Daily video production alone reportedly consumed $15 million in computational resources, with no evident route to financial sustainability.
  2. Despite initial signup constraints, user engagement deteriorated rapidly across successive months as novelty dissipated. The pattern mirrors typical behavior: after exchanging a few entertaining videos, interest evaporates.
  3. Harmful content surfaced with alarming frequency following the platform's debut. Documented cases ranged from fabricated military incidents in Gaza to merchandise designs referencing Epstein Island marketed toward children. This negative publicity created legal and reputational exposure for the company. Generating millions of videos daily exposed OpenAI to substantial liability risks and societal harms spanning child exploitation material to financial scams.

Meanwhile, established social networks face their own deluge of synthetic material—from trivial content like Shrimp Jesus to genuinely dangerous categories including child sexual abuse material, non-consensual intimate imagery, wartime disinformation, and fraudulent schemes.

Sora's collapse and the tepid reception to comparable systems like Meta's Vibes demonstrate a crucial lesson: audiences possess finite tolerance thresholds and ultimately abandon platforms through their choices. The evidence suggests that platforms built exclusively around synthetic content harm both public welfare and financial performance.

A learning moment

The moment has arrived—overdue, in fact—for technology companies, whether emerging AI specialists or established social media operators, to address these challenges systematically. Drawing on extensive academic and professional experience, several actionable measures exist for immediate implementation.

For AI content generators

  1. Embed content credentials (C2PA) into all AI-generated material. These cryptographically-verified labels communicate information about content origins to viewers and hosting platforms.
  2. Incorporate imperceptible watermarks across all output as supplementary verification, facilitating later identification (such as Google's SynthID approach).
  3. Reinforce safeguards governing user inputs and generated outputs to obstruct creation of prohibited or dangerous material.

For social media platforms

  1. Recognize content credentials and watermarks, clearly marking all synthetic material on their services using a labeling system analogous to nutritional information on packaged food—providing information without making value judgments.
  2. Furnish users with mechanisms to exclude synthetic content from their feeds, mirroring existing controls for blocking accounts or filtering content categories.
  3. Construct platforms with the foundational purpose of facilitating genuine human interaction rather than extracting monetary value from every unit of user attention.

For US policymakers

  1. Establish mandatory implementation of the safeguards and guardrails outlined above.
  2. Clarify that Section 230 protections do not extend to harms generated by AI systems themselves, distinguishing them from user-contributed content eligible for statutory protection.
  3. Enact federal legislation—following Denmark's recent precedent—establishing consumer rights to their own likenesses as a mechanism against fraud, disinformation, and non-consensual intimate imagery.

Should established social platforms and the broader technology industry disregard Sora's demise and mounting public frustration with synthetic content proliferation, they risk duplicating OpenAI's expensive miscalculation.

Source: Tech Policy Press