Editorial: Regulation and courts are moving faster than you think. Today’s pick of stories shows lawmakers, judges and cities pushing back—on opaque subscriptions, tax breaks for AI, unlicensed training data and algorithmic account takedowns. These aren’t niche policy scraps; they change how platforms operate and how creators, consumers and startups get paid.
In Brief
NYC makes cancellation one click away
Why this matters now: New York City’s click-to-cancel rule forces companies doing business with New Yorkers to let people cancel subscription services as easily as they sign up, immediately changing how subscription flows must be designed for millions of customers.
New York City’s Department of Consumer and Worker Protection has put a local rule into effect requiring online subscriptions sold to New Yorkers to be cancelable with the same ease as signup, and it creates a portal to report bad actors, with penalties for violations. The rule is a direct response to the common consumer pain of services hiding cancellation behind phone calls, long forms, or “call to cancel” hoops. As Mayor Zohran Mamdani put it, “If a company can take your money with one click, you should be able to get your money back with one click.” The Verge has the details on enforcement mechanics and the new complaint portal.
Policy note: this fills a gap left by a vacated federal rule and could become a model for other cities or states trying to curb bait-and-switch subscription practices. Companies that rely on complex cancellation flows will need to update UX and billing systems quickly if they want to keep serving New York customers.
Read more about the rule at The Verge.
Elizabeth Warren probes tech tax breaks tied to AI
Why this matters now: Senator Elizabeth Warren’s inquiry asks whether Amazon, Google, Meta and Microsoft are using tax credits and depreciation rules to shield roughly $19 billion in deductions related to AI investments—while broader changes to the tax base driven by AI may be costing the Treasury billions.
Sen. Warren has opened a probe into how big tech claims R&D credits, accelerated depreciation and other incentives as they pour money into AI infrastructure and compute, raising the political and fiscal stakes of the AI boom. The investigation frames the issue as a question of fairness: are incentives meant to spur innovation being used as permanent subsidies for already‑profitable firms? Media coverage ties the probe to broader concerns that AI-driven tax dynamics could be draining federal revenue. The Moneywise writeup outlines the scope of the request and frames the Congressional interest.
What to watch: companies under scrutiny will likely highlight the job-creation and technical investment side of their spending; policymakers will push for tighter definitions of eligible credits if public pressure grows.
Read the Moneywise summary here.
Deep Dive
A court rules training AI on editorial work can fail fair use
Why this matters now: A federal appeals court said that using a publisher’s editorial headnotes to train an AI legal-research product can violate copyright when the trained output competes with the original service—creating immediate licensing risk for AI firms that scrape curated content.
A recent appeals-court opinion—covered in TechSpot—centered on Ross Intelligence and Westlaw. The court found that Ross had used Westlaw’s editorially created headnotes in ways that could "substitute for Westlaw’s service." The opinion bluntly observed, > "In truth, this is no more than an ordinary copyright case," framing the issue around the traditional copyright concern of market harm rather than a novel AI exception.
Why the ruling matters: fair use is a flexible, four-factor test where one key factor is whether the new use harms the market for the original work. The court’s analysis emphasized that copying editorial labor (the curated, value‑added pieces that publishers sell) and using it to build a competing product tips the balance away from fair use. For AI builders, that distinction—between raw crawling of public-domain or user-generated text and ingesting paid, curated editorial content—matters a lot.
Practical implications:
- Startups and larger labs that train models on scraped news, headnotes, or other curated editorial content now face a clearer risk of copyright suits and possible injunctions or licensing demands.
- Business-to-business AI products that effectively replace subscription services (legal research, financial data terminals, etc.) are especially exposed because courts look at “market substitution” in the fair-use calculus.
- This decision doesn’t ban all unlicensed training, but it raises the bar for defensible datasets and increases the commercial value of licensing agreements with publishers.
Tech and legal teams should reassess training datasets now: build clear provenance audits, prefer permissive or public-domain sources where possible, and budget for licensing deals when editorially curated material is valuable to the product. Expect more case-by-case litigation and, likely, negotiated licensing markets between content owners and AI companies.
Read the TechSpot story on the ruling.
Overnight, Meta wiped her accounts; AI customer service blocked the appeal
Why this matters now: A creator lost Facebook and Instagram accounts overnight after automated enforcement, then found Meta’s AI-driven appeal flows refused to reach a human—highlighting how automated moderation can erase businesses and livelihoods without meaningful recourse.
A CBC feature follows a creator whose years of audience-building and income vanished when Meta’s systems deleted her accounts. When she sought help, Meta’s automated customer-service tools did not escalate to a human reviewer, leaving the creator effectively locked out. The New York Times framed the emotional and economic impact: the creator’s husband said it felt like "an actual physical business, and somebody just shuts down your office or your storefront." That quote captures why this is more than a platform annoyance.
Two trends converge here. First, platforms increasingly use automated classifiers to detect policy violations; second, they scale automated appeals or chatbots to triage complaints. Automation is efficient but brittle: classifiers can overreach, and appeal flows designed to filter low-priority cases end up trapping people with legitimate disputes. The lack of transparency compounds the harm—Meta does not publish automation error rates or the portion of takedowns resolved by humans, making it hard for regulators or researchers to measure systemic risk.
What this means for creators and small businesses:
- Build out backup channels and audience ownership (email lists, direct commerce endpoints, multi-platform presence). Platform followers are not the same as owned customers.
- Document identities, receipts, and business evidence proactively; it can shorten reinstatement when disputes are heard.
- Lobbying and regulatory pressure are likely to build: experts and rights groups now push for mandatory notice-and-appeal paths with human review for high-stakes action.
This case adds momentum to calls for clearer appeals guarantees and reporting on automation error rates. Platforms will face increasing political pressure to design appeal systems that can escalate to humans for decisions with real economic consequences—otherwise policymakers are likely to impose rules.
Read CBC’s coverage of the creator’s experience.
Closing Thought
Two takeaways tie these threads together: first, law and policy are catching up with the tech stack—the courts are policing training data, cities are rewriting consumer billing UX, and Congress is scrutinizing tax incentives for AI. Second, the practical space between “innovation” and “harm” is narrowing: businesses and builders must design with legal, fiscal and human-review realities in mind, not later retrofits. For creators and startups, that means both technical diligence (dataset provenance, simpler cancellation flows) and hard-nosed planning (audience ownership, legal budgets).