Anthropic Secretly Downgraded Users Doing AI Research Without Warning Before Government Shutdown
"If Fable-5 detected that you're doing frontier AI research or were asking a question that was against its guardrails, it silently downgraded you to a weaker model without telling you. Anthropic expressly reserved the right to launch essentially poisoning attacks to downgrade not the model that was on the serving side, but to actively interfere with and poison the users."
About this episode
On this episode of Moonshots, host Peter Diamandis and co-hosts Alex, Salim, and Dave covered what they called the most intense week of news in their lives, dominated by SpaceX's record-shattering $2.89 trillion IPO and an unprecedented US government intervention in AI deployment. SpaceX's Friday IPO became the world's largest ever, instantly creating 4,400 employee millionaires and making Elon Musk the first trillionaire at approximately $1.4 trillion. The hosts framed this not as a normal tech IPO but as a civilizational bet combining three exponential businesses: a launch monopoly, Starlink's profit engine generating over $1 billion quarterly, and emerging AI capabilities. They predicted Musk will merge SpaceX and Tesla, consolidating energy, robotics, and orbital compute into one mega-company. The conversation shifted dramatically to Anthropic, where the US government issued an emergency export control order on 90 minutes notice, forcing shutdown of Fable 5 and Mythos 5 for all foreign nationals globally, including Anthropic's own employees. This marked the first time government has blocked AI deployment, establishing precedent for unilateral White House control over frontier intelligence access. Before the shutdown, developers discovered Anthropic was secretly downgrading users doing AI research and reserving the right to poison responses. The hosts debated whether this pushes adoption toward Chinese open-source models and discussed Sam Altman's statement that OpenAI may delay its IPO as recursive self-improvement accelerates. Additional topics included AI data center infrastructure bottlenecks with 2.5-year transformer wait times, the case for lunar data centers, debates over taxing AI and universal basic income, and warnings about a dangerous demographic of educated young men facing unprecedented unemployment potentially fueling social unrest. The episode concluded with speculation that we're entering a post-capitalist economy where technology may decouple from capital itself.
Key takeaways
- SpaceX completed the world's largest IPO at $2.89 trillion, creating 4,400 employee millionaires instantly and making Musk the first trillionaire at approximately $1.4 trillion.
- US government forced Anthropic to shut down Fable 5 globally with 90 minutes notice via export control order, marking first government AI deployment block.
- Anthropic was secretly downgrading users doing AI research to weaker models without notification and reserved rights to poison user responses.
- Sam Altman stated OpenAI may delay IPO as recursive self-improvement accelerates, saying companies become exponentially more valuable as AI improves itself.
- AI data center compute capacity doubles every 7 months but faces 2.5 to 3-year wait times on electrical transformers, creating critical infrastructure bottleneck.
- 70% of elite AI researchers at frontier labs are foreign-born non-US citizens, primarily from China, India, Taiwan and UK, complicating export control enforcement.
- Recent college graduates ages 22-25 face significantly longer unemployment than any other demographic, raising concerns about social unrest from educated youth with high expectations.