Can AI Avoid the Enshittification Trap? (www.wired.com)

A writer’s pleasant experience getting a restaurant recommendation from GPT-5 prompts a broader worry: will powerful AI platforms repeat the “enshittification” arc Cory Doctorow describes—starting user-first, then privileging business partners, and finally extracting value from users? The piece highlights current signals that make that plausible: AI firms exploring ad products, integrations like OpenAI’s Walmart deal, and sponsored-result programs such as Perplexity’s paid placements (albeit labeled). Given the colossal capital outlays and concentration—companies planning to spend hundreds of billions—there’s strong incentive to monetize aggressively through advertising, paid tiers, changing fee structures, or repurposing user data for model training. GPT-5 itself reportedly agreed the framework maps “disturbingly well” onto AI if incentives go unchecked. For the AI/ML community this matters because trust, transparency, and incentive design are as consequential as model quality. Technical implications include opaque “black-box” models that can hide monetized biases or covert promotions, harder-to-detect shifts in objective functions as business incentives change, and consolidation that limits reproducibility and independent auditing. Practical defenses include better provenance and labeling for recommendations, explicit ad/sponsored markers, model cards and audits, privacy-preserving training (DP, federated learning), open-source alternatives, and regulatory scrutiny of dominant platforms. The debate underscores that engineering progress must be paired with governance and business-model choices to preserve AI’s value to users.