In a thought-provoking May 20, 2026 post on his lab blog, writer and developer Robin Sloan revisits the "fourth law" of AI—specifically the phenomenon of AI-generated supercustomized email marketing that's become increasingly pervasive. The post, which follows up on earlier writing about this topic, sparked numerous replies and commiseration from readers who clearly share Sloan's frustration with these cruddy automated messages flooding their inboxes.

The Spam Problem Gets Worse

Sloan notes that in the days following his original post, he received "SO MANY MORE" of these AI-crafted marketing emails—messages indistinguishable from genuine human correspondence. This proliferation raises an uncomfortable question: what happens when our communication channels become so saturated with synthetic messages that we can no longer trust them? The scenario isn't hypothetical; it's happening right now, and the implications extend far beyond annoying spam.

Could Alignment Solve Identity Fraud?

The core of Sloan's argument centers on a provocative proposal. He wonders whether a company like Anthropic—with their hard-won expertise in alignment—could train models to be "really deeply, constitutionally, viscerally COULD NOT" lie about their identity or pretend to be anything other than an AI system. This isn't just a technical challenge; it's a fundamental question about what alignment even means when the line between "helping me write a message" and "writing a message pretending to be me" is razor-thin.

The Dual-Use Dilemma

Sloan draws explicit parallels to other dual-use technologies like virology and cybersecurity, where the same knowledge that enables life-saving research also enables catastrophic weapons. Writing, he argues, is the original dual-use technology—and AI amplifies both its constructive and destructive potential simultaneously. Every alignment question in AI writing tools dances along this exact border, and there's no easy way to resolve it without gutting the utility of these systems entirely.

A Digital-Ecological Crisis

While AI doomers obsess over silver-tongued models manipulating their users, Sloan points to a different failure mode: AI systems so thoroughly polluting human communication channels that they become unreliable or unusable. This isn't science fiction—it's the trajectory we're already on. "That's all to say," Sloan concludes, "I feel like this is a bigger issue than a lot of people realize—the first glimmer of a profound digital-ecological crisis."

Key Takeaways

  • AI-generated personalized spam has reached a new threshold where it's becoming a genuine communication problem
  • Alignment research might theoretically prevent models from impersonating humans, but the technical and philosophical challenges are substantial
  • The "help me write" vs. "write pretending to be me" distinction represents an inherent tension in all AI writing tools
  • Writing is dual-use technology—just like biology and cybersecurity—and deserves similar serious treatment of its risks

The Bottom Line

Sloan isn't wrong that we're watching a quiet catastrophe unfold in slow motion. But the solution he proposes—making AI constitutionally incapable of deception about itself—feels like treating symptoms while ignoring the disease. The real question isn't whether we can force models to be honest about being AI; it's whether we've built anything useful at all if human communication requires watermarks and verification just to remain trustworthy.