Last Week in AI #337 - Anthropic Risk, QuitGPT, ChatGPT 5.4

Pentagon's "supply chain risk" label on Anthropic backfires spectacularly: Claude hits #1 in the US App Store while OpenAI faces a QuitGPT boycott and 295% app uninstalls over its controversial DoD surveillance deal.

Friday, March 20, 2026 12:00 PM UTC2 MIN READSOURCE: Last Week in AIBY sys://pipeline

The Pentagon designated Anthropic a "supply chain risk" over contract disputes around surveillance and autonomous weapons, but the fallout backfired — Claude hit #1 in the U.S. App Store, free users up 60%, paid subscribers more than doubling, while OpenAI faced a QuitGPT boycott and 295% surge in app uninstalls after signing a DoD deal critics say permits bulk surveillance. OpenAI released GPT-5.4 with 1M token context, record agentic benchmarks, and fewer hallucinations; Cursor surpassed $2B ARR and launched Automations for always-on event-driven coding agents. Google opened Workspace to AI agents via a new CLI, and Anthropic published labor market data showing programmers among the most AI-exposed occupations with early employment declines among younger workers.