Meta Own Product
Tennessee v. Meta Opens With 'Basically Pushers,' AI Productivity's Uncomfortable Counterevidence, and the 'What If We're Wrong' Test
State prosecutors opened the Tennessee trial against Meta by showing jurors internal company communications in which a Meta employee described Instagram as 'a drug' and called colleagues 'basically pushers.' The messages are not hacked documents — they are internal Slack messages and emails that Meta's own employees wrote. The product liability framing of the suit is distinct from earlier cases: rather than arguing negligent content moderation, Tennessee contends that Instagram was designed defectively, that its recommendation algorithm, notification architecture, and engagement loops are inherently harmful to adolescent users. A coordinated federal suit making the identical 'defective product' argument against both Discord and Meta draws a deliberate analogy to tobacco litigation, where internal documents eventually showed industry awareness of harms the companies publicly denied for decades. If courts accept the product liability framing, Section 230 immunity — which shields platforms from liability for user-generated content — does not apply, because it does not cover a company's own design choices.
Meta faces a separate content moderation failure: the company reportedly ran 7,600 AI-generated 'nudify' advertisements sourced through a Chinese advertising partner. The ads were paid placements that Meta's review systems accepted, raising questions about oversight of international advertising partners operating under different content standards.
The Morgan Stanley finding that AI adoption is measurably boosting corporate profits represents the current consensus view in the business press. But a survey by Adaptavist offers a pointed challenge: 38 percent of workers said they would remove generative AI entirely from their workflows, and nearly half reported spending more time verifying AI output than the tools save them. Both data points can be simultaneously true if the profit gains Morgan Stanley identified are driven primarily by headcount reduction rather than genuine output improvement — meaning AI is functioning as labor substitution wearing a productivity costume. The profits are real; the causal story may not be.
Three specific signals would confirm or refute that hypothesis: customer satisfaction scores declining at companies aggressively deploying AI while profit margins hold; elevated turnover among high-skilled employees best positioned to catch AI errors, as verification burden becomes unsustainable; and a quarter of Morgan Stanley-style profit improvement followed by a spike in error rates, recalls, or customer complaints. Investors treating AI adoption as straightforwardly bullish for enterprise productivity should, at minimum, be tracking net promoter scores at the largest AI deployers over the next two quarters.