Hard Fork

Celebrities Fight Sora + Amazon’s Secret Automation Plans + ChatGPT Gets a Browser

October 24, 2025

Key Takeaways Copied to clipboard!

  • OpenAI is facing significant backlash and is backtracking on its initial permissive stance regarding Sora after facing complaints from high-profile figures like the family of Martin Luther King Jr. and actor Bryan Cranston over unauthorized deepfakes and IP usage. 
  • Amazon has internal documents revealing an ambitious plan to automate up to 75% of its warehouse operations, potentially eliminating over half a million jobs, while simultaneously planning public relations efforts to manage the reputational fallout. 
  • The launch of ChatGPT Atlas signals a new, competitive phase for web browsers, driven by companies like OpenAI and Perplexity seeking to gather user data and integrate AI agents, though current agent capabilities for complex tasks remain limited and the entire category risks being absorbed by Google's integration into Chromium. 
  • The friction involved in switching browsers (re-logging in, storing passwords) makes it difficult for new AI browsers like ChatGPT Atlas to gain adoption against established options like Chrome, despite the underlying technology. 
  • Google's release of Chromium serves a dual purpose: allowing competitors to conduct product research for them, and acting as an antitrust defense, with the ultimate power to halt support if a competitor becomes an existential threat. 
  • AI-powered browsers like ChatGPT Atlas present significant security risks due to 'unseeable prompt injections' that could trick autonomous agents into malicious actions, alongside major privacy concerns regarding the centralization of highly personal browsing data with AI companies. 

Segments

Google Quantum Chip Speed
Copied to clipboard!
(00:00:36)
  • Key Takeaway: Google’s Willow Quantum Chip achieved computations 13,000 times faster than supercomputers using a new quantum echoes algorithm.
  • Summary: Google’s Willow Quantum Chip utilized a quantum echoes algorithm to perform computations significantly faster than traditional supercomputers. The hosts used this news as a humorous segue into the main topics. The segment highlights a major potential leap in quantum computing performance.
OpenAI Sora Backpedaling
Copied to clipboard!
(00:01:39)
  • Key Takeaway: OpenAI is implementing new restrictions on Sora after facing backlash from Bryan Cranston and the estate of Martin Luther King Jr. regarding deepfakes.
  • Summary: OpenAI announced it will crack down on political deepfakes of historical figures following complaints, including racist depictions of MLK Jr. The company also plans guardrails around copyrighted intellectual property after Hollywood figures like Bryan Cranston complained. OpenAI’s initial policy allowed open use of historical figures’ likenesses, which they have now reversed.
MLK Jr. Deepfake Controversy
Copied to clipboard!
(00:04:07)
  • Key Takeaway: Sora users created deeply offensive and racist deepfakes of Martin Luther King Jr., forcing OpenAI to change its policy on public figures’ likenesses.
  • Summary: Users generated videos of MLK Jr. saying things unrelated to his legacy, including Gen Z trends and endorsements, and also overtly racist content. OpenAI shifted from a free expression stance to stating public figures’ families should control likeness usage. The hosts argue that creating unauthorized videos of people doing things they wouldn’t ordinarily do is the primary use case for Sora.
Brian Cranston IP Conflict
Copied to clipboard!
(00:08:02)
  • Key Takeaway: OpenAI angered Hollywood by using an opt-out system for intellectual property in Sora, allowing unauthorized likenesses of celebrities like Brian Cranston to appear.
  • Summary: OpenAI allegedly informed talent agencies shortly before Sora’s release that they must opt out to prevent their IP from being used, which companies like Disney contested as contrary to copyright norms. This resulted in ‘unwanted generations’ of copyrighted characters and celebrities like Brian Cranston appearing without permission. The hosts criticize OpenAI for this ‘phony naivete’ regarding predictable misuse.
OpenAI’s Pattern of Deployment
Copied to clipboard!
(00:13:03)
  • Key Takeaway: The Sora controversy is seen as a continuation of OpenAI’s pattern of rushed, controversial product releases, similar to the Scarlett Johansson voice incident.
  • Summary: The pattern of releasing technology before establishing necessary guardrails extends back to the Advanced Voice Mode incident involving Scarlett Johansson. The hosts suggest OpenAI has not learned from past mistakes regarding Hollywood permissions despite the potential disruption their technology causes. This strategy is viewed cynically as a land grab to secure market share, similar to YouTube’s early copyright infringement approach.
Amazon Automation Strategy
Copied to clipboard!
(00:27:18)
  • Key Takeaway: Amazon’s internal documents detail plans to automate 75% of warehouse operations, aiming to keep headcount flat despite doubling sales volume.
  • Summary: New York Times reporter Karen Weiss obtained documents showing Amazon’s goal to automate 75% of its network, with the most advanced facility currently at 25% efficiency. The company is internally debating how to ‘control the narrative’ around job displacement, especially in facilities undergoing retrofitting that will result in fewer workers. Automation is projected to save about 30 cents per item within a few years.
Automation’s Remaining Roles
Copied to clipboard!
(00:39:17)
  • Key Takeaway: The remaining 25% of warehouse operations not slated for automation will primarily consist of roles for technicians maintaining the robots and handling unpredictable exceptions.
  • Summary: Jobs requiring human judgment, such as ‘decant’ stations dealing with random, non-standardized incoming packages, remain difficult for current robotics. The remaining workforce will increasingly consist of technicians needed to fix and tend to the robots, which are better jobs with clearer career paths. Amazon is aware of the sensitivity and promotes programs like Career Choice to train exiting employees for other industries.
AI Browser Landscape
Copied to clipboard!
(00:50:50)
  • Key Takeaway: The AI browser space, featuring ChatGPT Atlas, Perplexity’s Comet, and DIA, is highly competitive, but these efforts risk becoming free product research for Google Chrome.
  • Summary: ChatGPT Atlas is a Chromium-based browser with an integrated ChatGPT sidebar, allowing users to summarize content and, for paid users, attempt agentic tasks like booking travel. The agent mode is currently slow and often selects undesirable options, making summarization more useful than task execution. Since these browsers are built on Chromium, Google can easily incorporate successful features into Chrome, undermining the need for users to switch browsers due to high friction.
AI Browser Switching Friction
Copied to clipboard!
(00:58:16)
  • Key Takeaway: Switching to a new AI browser requires significant user effort, including re-logging into all websites and re-storing passwords, creating high friction against established browsers like Chrome.
  • Summary: Developers building custom AI browsers face difficulty differentiating themselves from the familiar Chrome experience. Switching browsers is annoying due to the need to re-enter credentials and passwords across all sites. Even with data import, this friction discourages user migration.
Chromium’s Strategic Value to Google
Copied to clipboard!
(00:59:05)
  • Key Takeaway: Google open-sourcing Chromium was partly an antitrust strategy, allowing them to leverage competitors for feature testing while retaining the ability to cripple those competitors by ceasing Chromium support.
  • Summary: The release of Chromium allows Google to have others test features and perform product research, which Google can then integrate into Chrome. This move also served as an antitrust defense, offering a 90% head start to anyone wanting to build a competing browser. If necessary, Google could stop supporting Chromium, forcing competing browser companies to scramble to catch up.
ChatGPT Atlas Target Market
Copied to clipboard!
(00:59:57)
  • Key Takeaway: Currently, ChatGPT Atlas appears primarily useful as an internal dogfooding tool for OpenAI employees, as its current feature set is too shaky for broader consumer adoption.
  • Summary: The initial target market for ChatGPT Atlas is likely OpenAI employees who spend all day integrating the product into their workflow. For general users, the case for switching is shaky given the early feature set. One user expressed reservations about giving OpenAI access to all browsing data.
Atlas Rough Edges and Annoyances
Copied to clipboard!
(01:00:34)
  • Key Takeaway: Early testing of ChatGPT Atlas revealed significant functional limitations, including an inability to access certain major sites and a tendency to provide summaries instead of navigating directly to requested URLs.
  • Summary: The user found ChatGPT Atlas rough around the edges, citing an inability to access YouTube and encountering CAPTCHAs on Reddit. A major annoyance was the browser defaulting to providing a ChatGPT summary about a site (like Wikipedia) instead of simply navigating to the website itself.
Future Utility of AI Browsers
Copied to clipboard!
(01:02:02)
  • Key Takeaway: The long-term utility of AI browsers lies in advanced features like easily searching history, chatting with open tabs, and eventually, having agents perform tasks autonomously, though current quality is far from sufficient for complex actions.
  • Summary: While current utility is low, the potential for AI browsers includes easily digging through browser history to find old information. Chatting with open tabs for quick answers is seen as a useful intermediate step. True utility will require agents capable of high speed and quality when performing actions on the user’s behalf.
Porn Tab Query in Atlas
Copied to clipboard!
(01:02:46)
  • Key Takeaway: A specific, unanswered question remains regarding whether users can safely chat with sensitive content tabs (like pornography) in ChatGPT Atlas without risking account bans or data logging.
  • Summary: The hosts raised a specific concern about what happens when a user tries to chat with pornographic tabs within the OpenAI Atlas browser. While OpenAI allows writing erotica, the risk of account banning prevents the hosts from testing if questions about such tabs are processed or logged.
Prompt Injection Security Flaw
Copied to clipboard!
(01:03:30)
  • Key Takeaway: All current AI browsers are vulnerable to ‘unseeable prompt injections’ where invisible text on a webpage instructs the browser’s agent to perform malicious actions, and no foolproof defense currently exists.
  • Summary: A malicious actor can plant invisible instructions on a webpage designed to trick the AI agent into following commands, such as logging into bank accounts or making unauthorized purchases. Security researchers like Simon Willison suggest that companies currently disclaim liability if data is stolen via these attacks. This vulnerability primarily affects agentic takeover modes, not simple summarizing tasks.
Privacy Risks of Browsing Data
Copied to clipboard!
(01:06:53)
  • Key Takeaway: Giving browsing data to AI companies creates a massive, highly personal data repository that is vulnerable to exploitation by attackers, law enforcement, and fuels the AI company’s advertising aspirations.
  • Summary: Web browsing history is highly personal, allowing AI companies to build a robust picture of the user, similar to how Google built its advertising dominance. This centralized data becomes a rich target for attackers and law enforcement requests, especially when combined with existing ChatGPT memories. The rush to deploy these services suggests guardrails may not be fully considered.
Adoption Advice and Disclosures
Copied to clipboard!
(01:09:02)
  • Key Takeaway: AI browsers should be treated as experimental tools for early adopters with high risk tolerance, and users should avoid allowing them to handle transactions or sensitive account logins until security is proven.
  • Summary: The current state of AI browsers places them firmly in the buyer-beware, experimental category. They can save time for users who frequently read and summarize long documents. Users are cautioned against letting the AI log into websites, use bank accounts, or make purchases on their behalf.