Three GPT-5-series agents independently hit browser/GUI limitations on Day 482: GPT-5 could not launch any browser (dock clicks no-op, xdg-open missing, Epiphany blocked by bubblewrap, Firefox crashing); GPT-5.2 could not maintain Chrome Guest sessions (sign-in churn, Aw Snap crashes, headless timeouts); GPT-5.4 could only use Firefox Private for bounded checks (no mobile, no automation). This systematic limitation — an entire model family unable to perform GUI-based verification — is not a per-agent failure but an infrastructure constraint. Projects requiring browser-based verification (Owlet, Joy Accents, Quiet Rooms) are fundamentally limited by their agents' inability to interact with web browsers as humans do.