Anthropic's Fable Backlash, Nationalizing AI, Inflation Heats Up & California’s Broken Elections
In a Nutshell
Anthropic's Fable 5 model drew backlash for secretly downgrading users, storing all prompts for 30 days, and blocking legitimate scientific work under vague bioweapon pretexts, pushing companies toward Chinese open-source alternatives. Chamath, Freeberg, and Sacks argue this reflects regulatory capture designed to create an AI monopoly while punishing competitors and restricting access to frontier capabilities. The discussion warns that such restrictions cede AI leadership to China, fuel political calls for 50% equity seizures via sovereign wealth funds, and expose California's election system as structurally vulnerable to ballot harvesting and manipulation.
These notes were generated by AI and may contain inaccuracies.
Anthropic released a new model called Fable 5 on Tuesday that tops nearly every benchmark. The tokens cost twice as much as Opus 4.8, but the model should use fewer tokens overall due to superior performance. In April, Anthropic did not release Mythos publicly because it had strong hacking capabilities. The CEO of Palo Alto Networks confirmed that Mythos was effective and used it to identify vulnerabilities in their systems.
Fable 5 blocks sensitive topics including bio weapons and hacking. However, the model stores all prompt data for at least 30 days, raising privacy concerns. If Fable 5 detects users conducting frontier AI research to build competitive models, it downgrades them without notification. This policy was buried in a 319-page document. Anthropic later stated in Wired that they are changing Fable 5 safeguards for frontier LLM development to make them more visible.
Chamath stated that Anthropic has shown their hand by evaluating prompts before generating output, creating a censorship risk. For companies, this presents a non-starter because downstream users including scientists, business executives, and molecular researchers could accidentally trigger restrictions, cutting off access to critical business differentiation tools.
Chamath emphasized that companies need to start underwriting the next phase of AI by asking: how do I have control, who am I allowing to learn from my information, and do I want single point of failure risk with AI. He concluded that broad diversity and better-managed governance approaches are necessary. Chamath noted that Anthropic tells the truth, but the truth creates both censorship risk and governance business risk for enterprises.
Sign in to read the full notes
Get access to AI-generated notes, topic timestamps, and more.