Monologue: What's Going On At Anthropic?

In this monologue, host Zitron examines recent troubles at Anthropic, centering on Claude's subscription service, sudden rate limit impositions, security leaks, and the risks of using large language models (LLMs) like Claude to generate production code. He argues these issues reveal deeper problems in Anthropic's operations, including an unsustainable subsidized model and insufficient oversight of AI-written software.

Claude Subscriptions and Token Economics

Claude subscriptions cost $20, $100, or $200 per month, granting tiered access via rate limits like five-hour session caps and weekly quotas, with higher tiers for Opus, Anthropic's advanced model. These plans are heavily subsidized: users can burn thousands of dollars in underlying token costs—up to $2,500 or more on a $200 plan—without per-token billing, unlike API users who pay per million input/output tokens. This setup previously allowed extensive use equivalent to API features, but Anthropic has tightened limits to manage costs and approach profitability, as unchecked usage would require extreme restrictions.