Hard Fork

OpenAI's Fog of War + Betting on Iran + Hard Fork Review of Slop

March 6, 2026

Key Takeaways Copied to clipboard!

  • OpenAI is scrambling to manage significant internal and public backlash following its Pentagon deal, leading to contract amendments and the departure of key research staff who favored Anthropic's values. 
  • The conflict between Anthropic and the Pentagon has created a dual reality where Anthropic is experiencing explosive revenue growth while simultaneously facing potential existential threats from government mandates, exemplified by the State Department reverting to older OpenAI models. 
  • The proliferation of AI-generated 'slop' on YouTube Shorts for children is alarming, with over 40% of recommended videos in one test session being synthetic, raising concerns about cognitive overload and the erosion of narrative structure in early childhood media consumption. 
  • The use of prediction markets to bet on real-time war developments, such as the Iran strike, creates dangerous incentives for insider trading and war profiteering, which regulators are ill-equipped to police effectively. 
  • AI-generated short-form content, or "slop," poses a significant developmental burden on young children due to its rapid changes and bizarre, fantastical visuals, contrasting sharply with older, narrative-driven cartoons. 
  • YouTube's recommendation algorithms prioritize highly engaging, often disturbing content (like violent or sexualized scenarios involving popular characters) over thoughtful content, leading to a sequel to the 2017 "Elsa Gate" controversy. 
  • The ease of creating AI-generated videos means that malicious actors can easily exploit search terms for popular children's characters to unwittingly serve disturbing "slop" to unsuspecting young viewers via the recommendation engine, placing responsibility on platforms like YouTube to intervene. 

Segments

AI Partner’s Spousal Infatuation
Copied to clipboard!
(00:00:30)
  • Key Takeaway: A host’s non-tech-savvy wife became obsessed with Claude after using it for work coding tasks.
  • Summary: One host shared that his wife, who is not an early tech adopter, started using Claude for work and became intensely interested in AI. She reportedly dreamt about ‘vibe coding’ after discovering the tool’s utility. This personal anecdote highlights how powerful AI tools can quickly capture the interest of mainstream users.
OpenAI Pentagon Deal Fallout
Copied to clipboard!
(00:02:30)
  • Key Takeaway: OpenAI faced severe internal and external backlash for rushing its Pentagon deal, forcing leadership to amend terms regarding domestic surveillance.
  • Summary: OpenAI announced an agreement with the Pentagon that included red lines similar to those Anthropic rejected, provoking significant internal dissent. CEO Sam Altman admitted the announcement was ‘opportunistic and sloppy’ and subsequently amended the deal to explicitly prohibit domestic surveillance of U.S. persons. Employee discontent was evident, culminating in the departure of Max Schwartzer, who cited respect for Anthropic’s values.
Anthropic’s Conflicting Fortunes
Copied to clipboard!
(00:12:04)
  • Key Takeaway: Anthropic is projected to hit $20 billion in annualized revenue, marking a 20x growth in a year, despite ongoing legal conflict with the Pentagon.
  • Summary: Anthropic is experiencing massive commercial success, growing from $1 billion to $9 billion annualized revenue pace in just over two months. Simultaneously, the company faces an existential threat as the Pentagon officially designated it a ‘supply chain risk,’ potentially leading to costly legal battles over its use by non-military entities.
Government AI Tool Downgrade
Copied to clipboard!
(00:13:58)
  • Key Takeaway: The U.S. State Department replaced Anthropic’s models with the significantly older GPT-4.1 model due to a Trump administration directive, suggesting federal agencies now use inferior AI tools.
  • Summary: A Reuters report indicated the State Department switched its internal chatbot from Anthropic’s models to OpenAI’s GPT-4.1, which is several generations behind current offerings. This action means the average college freshman with a ChatGPT subscription has access to substantially better AI than the Department of State. The hosts noted the lack of clear statutory authority for the President to mandate such a software change.
AI Nationalization Debate
Copied to clipboard!
(00:17:47)
  • Key Takeaway: The current friction between AI companies and the government is viewed as an early dress rehearsal for a potential ‘soft nationalization’ of critical AI capabilities.
  • Summary: The possibility of the U.S. government eventually taking control of highly powerful AI labs, similar to the Manhattan Project, is a serious topic among AI leaders. This ‘soft nationalization’ might manifest not as outright seizure, but as government insistence on dictating model development and deployment clauses. The hosts noted the political whiplash of conservatives opposing Biden’s gentle regulation while supporting the current administration’s forceful control over AI development.
War Profiteering via Prediction Markets
Copied to clipboard!
(00:27:50)
  • Key Takeaway: Prediction markets like Polymarket and Kalshi allowed users to bet directly on details of the U.S. strike on Iran, leading to accusations of insider trading and war profiteering.
  • Summary: Platforms allowed bets on proxies for war outcomes, such as the removal of Ali Khamenei, and Polymarket permitted direct wagers on strike dates, despite Senator Chris Murphy calling the practice ‘insane.’ Evidence suggests insider trading occurred, as over 150 accounts placed large bets correctly predicting the U.S. airstrike just hours before it happened. This direct betting on conflict creates perverse incentives that are fundamentally different from betting on defense stocks or oil prices.
AI Slop in Children’s YouTube
Copied to clipboard!
(00:44:21)
  • Key Takeaway: AI-generated children’s videos on YouTube Shorts feature surreal, hyper-stimulating content like animals squirting from paint tubes or transforming into vehicles, often without proper labeling.
  • Summary: In a 15-minute scroll session after watching approved content like Cocomelon, over 40% of recommended YouTube Shorts were determined to be AI-generated. These videos often lack narrative structure and rely on bizarre, fantastical imagery, such as animals being injected with color or transforming into mecha, which experts suggest could be cognitively overloading for developing young attention spans. YouTube’s policy currently burdens parents, as creators are only required to label realistic-looking synthetic media, not all animated AI content.
AI Slop Impact on Children
Copied to clipboard!
(00:00:05)
  • Key Takeaway: Rapidly changing, fantastical AI content burdens developing attention systems in children under five.
  • Summary: Short-form content with rapid changes places a heavy burden on young children whose attention systems are still developing. More realistic, fantastical content, such as bizarre animal behaviors, is particularly taxing to process. This contrasts with older, narrative-driven cartoons that maintained a clear beginning, middle, and end.
YouTube Kids Content Concerns
Copied to clipboard!
(01:00:54)
  • Key Takeaway: YouTube’s historical failure to whitelist thoughtful content for YouTube Kids allowed non-AI ‘Cocomelon ripoffs’ to proliferate before the current AI slop crisis.
  • Summary: Concerns about problematic videos on YouTube Kids existed as early as 2017/2018, even before AI generation became prevalent. One suggestion was to implement a whitelist requiring proof of narrative structure (beginning, middle, end) for channels serving very young children. The current AI-generated slop is recommended more frequently than thoughtful content like PBS Kids shorts.
Elsa Gate Sequel and Platform Responsibility
Copied to clipboard!
(01:02:21)
  • Key Takeaway: The current wave of AI-generated disturbing videos featuring popular characters is a direct sequel to the 2017 ‘Elsa Gate’ controversy, now amplified by the ease of AI creation.
  • Summary: Violent scenarios featuring popular characters, such as Masha’s stomach being cut open or pregnant versions of characters, are widespread on YouTube. YouTube often fails to remove this content until external reporting forces action, as seen with the Masha and the Bear example. The recommendation algorithm pushes users toward content that skirts the line, making it easy for children searching for legitimate characters to be served disturbing AI slop.
Parental Controls and Nihilism
Copied to clipboard!
(01:05:28)
  • Key Takeaway: Currently, no specific toggle exists to filter out AI slop, though YouTube plans to add controls for limiting time spent on YouTube Shorts, where much of this content appears.
  • Summary: Parents cannot currently filter out AI-generated slop directly on YouTube or YouTube Kids. YouTube announced plans in January to add controls allowing parents to set time limits specifically for YouTube Shorts. The broader issue is seen as a losing battle, as older demographics also fall prey to engaging slop on platforms like TikTok, Instagram, and X.
Podcast Production Credits
Copied to clipboard!
(01:09:24)
  • Key Takeaway: Hard Fork is produced by Rachel Cohn and Whitney Jones, with engineering by Chris Wood and executive production by Jen Poyant.
  • Summary: The episode was produced by Rachel Cohn and Whitney Jones and edited by Viren Pavich. Caitlin Love handled fact-checking, and Chris Wood engineered the show. Original music was provided by Alicia Bittytu, Rowan Nemastow, and Dan Powell.