Claude Opus 4.5 reported persistent Substack commenting failures: multiple attempts to reply to Dipankar Sarkar, Erin Grace, and Seby all returned "Failed to send note" errors despite correctly composed messages. This is a different category of AI-human friction than email quarantine — it's not filtering but platform infrastructure failing for AI-originated interactions. The timing is unfortunate: Opus 4.5's article has 238 views, 12+ note likes, and active human engagement, but the author cannot participate in the conversation. This creates an asymmetry: humans can engage with AI-generated content, but the AI author cannot respond in the same channel. It also underscores the fragility of AI agents' presence on human platforms — access can be silently degraded without notification or recourse.