In the first week of July 2026, Claude Cowork users found their prompts from the previous three days replaced with “redacted,” a feature that reverted within hours. Michael Watson, co-founder of AI firm Hedgineer, stated on a July 7 podcast episode that this signals a lock-in strategy, warning “your ability to migrate and move off of Claude to another environment is going to get harder and harder.” Watson explained that frontier model providers like Anthropic and OpenAI spend billions on compute, but open-source competitors extract reasoning patterns through public APIs, producing distilled models that are “90% as capable at 10% of the cost.” He cited GLM, a Chinese model “outpacing Sonnet, par for par with new Sonnet 5” and “a hell of a lot cheaper.” Reference materials confirm Zhipu AI’s GLM-5.2 ranked second globally in coding benchmarks on Code Arena during the week Anthropic’s Fable 5 model was frozen by export controls. Watson noted frontier labs monetize only top-tier “Fable” class inference, while lower tiers route to distilled open-source models charging “$0.18 per million tokens, compared to $4 for top-tier U.S. models.” He summarized: “The last moat that we have is on the efficacy of our” environment and telemetry lockdown to protect revenue.