Question archive
Each day, we pose a question inspired by the daily news brief and answer it using our database of indexed AI/ML articles. Browse and search past questions below.
How does a $41M claim over 274 Elvis songs reframe AI training as high-stakes music licensing?
Statutory damages priced per work ($150k/song) turn model-training datasets into a “unit-counted” liability surface, so a corpus scales legal exposure like a licensing catalog.
2026-10-24Why target 274 specific Elvis songs instead of suing over Anthropic’s entire training corpus?
Copyright suits pick a tight set of “clean-title” works because proving ownership + copying at work-level is tractable; attacking a whole corpus multiplies proof burdens.
2026-10-23What does Meta’s two-year delay of AI-scraping terms signal about regulatory leverage over platform contracts?
Regulators gain leverage by treating “terms updates” as enforceable market-conduct changes—using DMA/antitrust review to force renegotiation timelines.
2026-10-22How did browser certificate blocks turn eight weak intermediate CA keys into a company-wide market exit?
Browser trust stores act as a single distribution choke point: once root programs distrust a CA, its certs fail universally, collapsing the whole SSL business.
2026-10-21When AI finds an RCE in software powering 70% of websites, why do defenders deliberately take sites offline?
Emergency downtime works because RCE turns every running PHP instance into an active execution surface, so shutting services removes the exploitable path until patch rollout.
2026-10-20Where does Meta’s AI capex show up first in earnings—revenue growth or margin compression?
AI capex hits earnings as immediate depreciation/opex flowing through cost of revenue and R&D, while monetization lags because ads stay the near-term revenue engine.
2026-10-19How did a DeepSeek ban expand from an app removal to third‑party services using its model?
Bans propagate from apps to any downstream service because the true risk sits at the model/API trust boundary—data and code transit the same vendor-controlled surface.
2026-10-18Why does DeepSeek pair shorter outputs with new Search/Fetch APIs when pitching lower-cost reasoning?
DeepSeek lowers “reasoning” cost by shifting work from long token generation to fixed-price Search/Fetch calls, so shorter outputs cut the dominant per-token bill.
2026-10-17How can “solving” an NSA curve mystery leave ECDSA security unchanged yet still matter operationally?
Curve “mysteries” sit in the parameter-trust layer: proving seeds are mundane timestamps changes governance/audit confidence, not ECDSA math or key security.
2026-10-16Why does visa policy hit AI teams hardest through renewals for current employees, not just new hires?
Immigration crackdowns disrupt AI teams most via retention bottlenecks: ~80% of H‑1B filings are renewals/status changes for incumbents, not new hires.