Documents unsealed in October 2026 in the consolidated OpenAI copyright MDL (S.D.N.Y., No. 1:25-md-03143) showed Microsoft directed employees to keep confidential a project supplying OpenAI with billions of scraped webpages (internally called 'Project Taxi') and a crawler built for OpenAI ('Project Mango'). Internal Microsoft and OpenAI documents showed both companies recognized their products were causing steep drops in referral traffic to publishers (as much as 83-93% for some sites) and risked creating a 'doom loop' that could gut the web's economics, while continuing the practices and keeping the scraping arrangement secret.
Documents unsealed September 17, 2026 in NYT v. OpenAI/Microsoft copyright litigation (SDNY, Judge Sidney Stein) quote Microsoft Director of Applied Science Dr. Brent Hecht internally describing the company's AI training data practices as 'an astonishing theft of unprecedented proportions' and 'perhaps the largest theft of labor in human history,' warning a ruling favoring Microsoft would make 'a complete mockery' of fair use. Microsoft's own internal data cited in the filings showed Bing Copilot cut click-through rates to publisher sites by 87-93% for the New York Times, 83-91% for Daily News, and up to 94% for Ziff Davis compared to traditional search. Other internal documents warned the practice had started a 'doom loop' and that 'LLMs are a product that destroys its supply chain.' Microsoft said Hecht's comments reflected one employee's individual view, not a legal analysis.
negligent
A 404 Media investigation demonstrated that anyone could hijack a real, verified artist's Spotify page by generating an AI song (via Udio), submitting it through distributor DistroKid under the artist's existing name, and falsely checking boxes claiming authorization -- with no verification performed by DistroKid or by Spotify (or Apple Music, Tidal, Amazon Music, Deezer). The reporter's fabricated track appeared on Brooklyn band Lathe of Heaven's official Spotify page within a day, on 2026-09-10. Musician Odette Child had separately compiled and forwarded to Spotify a list of 300+ AI-generated impostor songs -- affecting artists including Taylor Swift and deceased musicians -- which Spotify acknowledged receiving but did not act on. Spotify said reliable AI-detection remains 'a technically unsolved problem' and pointed to an in-beta 'Artist Profile Protection' opt-in review tool, first announced in March 2026 and still not generally available. Because Spotify's royalty payout has a roughly 8-week delay, scammers often collect payment before an impostor track is identified and removed, at return-on-investment estimated above 7,000%.
Anthropic and U.S. national security agencies (NSA, FBI, CISA) reported that Alibaba conducted a systematic 'distillation' campaign extracting Claude's reasoning transcripts and internal model-development data to train its Qwen models. Between May and July 2026, Alibaba-linked accounts made 151+ million unauthorized exchanges with Claude, peaking at nearly 3 million per day from 3,500+ fraudulent accounts using residential proxies, disposable emails, and virtual cards. Exchanges included sensitive information from individual users, multinational companies, and state-affiliated actors. On September 8, 2026, NSA/FBI/CISA issued a joint statement naming Alibaba (alongside DeepSeek, Moonshot AI, MiniMax, StepFun, and Z.AI) for systematically extracting proprietary AI knowledge from American firms at industrial scale since at least 2024.
reactive
On August 11, 2026, Spotify announced an 'AI Personas' badge system labeling artist profiles whose identity 'may be AI-generated and does not represent a real person.' Two badge variants were introduced: a voluntary 'AI Persona' self-disclosure tag and a 'Likely AI Persona' tag applied by Spotify's audit team using human reviewers plus AI investigative tools, with an appeals process for flagged artists. Music from labeled AI Persona profiles is excluded from personalized recommendations by default, and listeners will soon be able to report suspected AI personas. The move follows EU AI Act content-labeling rules for AI-mimicked voices that took effect August 2, 2026, and industry pressure after incidents such as the AI band 'Velvet Sundown' gaining a Spotify verified badge and over a million streams in 2025.
A stockholder derivative lawsuit filed July 31, 2026 in the US District Court for the Northern District of Illinois (Berliner v. Huang et al.) accused Nvidia's CEO Jensen Huang, CFO Colette Kress, and the board of breaching fiduciary duty by knowingly training NeMo Megatron and other language models on pirated 'shadow library' datasets (Anna's Archive, LibGen, Sci-Hub, Z-Library, Books3/SlimPajama), including material Nvidia allegedly sourced directly from Anna's Archive in 2023 despite an internal warning about copyright issues. The complaint also alleges Nvidia's Magpie TTS, FUGATTO, PersonaPlex and other voice-synthesis models extracted biometric voiceprints from hundreds of thousands of hours of speech recordings without notice or consent, in violation of the Illinois Biometric Information Privacy Act, and that proxy filings misrepresented data-sourcing compliance.
Hachette Book Group, Cengage Learning, Elsevier, and author Scott Turow filed a class action against Google LLC on July 10, 2026, alleging Google copied millions of copyrighted works to train its Gemini AI models. The complaint alleges Google misused books obtained through the Google Books program (provided only for search functionality), downloaded pirated works from shadow libraries and behind paywalls, and intentionally removed or altered copyright management information to conceal the source of the training data. An internal Google document cited in the complaint reportedly acknowledged the practice could be 'highly problematic' with potential exposure of '$10Bs-$100Bs in potential fines.'
After the French Senate unanimously adopted the cross-party Darcos-Ouzoulias bill in April 2026 - which would have reversed the burden of proof in AI copyright-infringement cases, requiring AI companies to prove they did not use copyrighted works rather than creators proving they did - Mistral AI's public affairs director Audrey Herblin-Stoop told the National Assembly's economic affairs committee the bill was a 'litigation bonanza' with 'impossible proof' requirements, warning 'if we face lawsuits, we must be clear: we will struggle to raise funds.' Mistral then conducted an intensive lobbying campaign targeting parliamentary group leaders, including a May 7, 2026 meeting between CEO Arthur Mensch and National Assembly President Yael Braun-Pivet. On May 12, 2026, the bill was kept off the Assembly's agenda, despite a joint call for passage from 81 cultural and creators' organizations and a petition signed by over 25,000 people.
compelled
In Muoki & another v Safaricom PLC & another [2026] KEHC 6326 (KLR), the Kenyan High Court found that Safaricom copied software developer Peter Nthei Muoki's 'M-Teen Mobile Wallet USSD Code' to launch its own nearly identical 'Manage Child Account' / 'Manage Junior Account' feature (USSD code *334#) without license or compensation. The court awarded KES 1,400,067,000 (~$10.8M) in general damages plus interest, and ordered Safaricom to pay an ongoing royalty of 0.5% of gross M-Pesa revenue annually, starting from the financial year ended March 31, 2025, for as long as it operates the infringing or substantially similar functionality.
negligent
Anysphere launched 'Composer 2' inside its Cursor code editor on March 19, 2026, marketing it as frontier in-house coding intelligence without disclosing its base model. Within days, a developer identified the underlying model fingerprint as Moonshot AI's open-weight Kimi K2.5, and Moonshot's head of pretraining publicly accused Cursor of violating Kimi's modified-MIT license (which requires prominent 'Kimi K2.5' UI attribution and licensing fees above $20M in monthly revenue or 100M MAU -- thresholds Cursor's roughly $2B annualized revenue clearly exceeds) without paying fees. Cursor VP Lee Robinson and co-founder Aman Sanger acknowledged Composer 2 started from the open-source Kimi base, with Sanger calling the lack of disclosure in Cursor's launch blog post 'a miss.' Moonshot separately stated Cursor's use came through an authorized commercial partnership with Fireworks AI, muddying the licensing-violation claim.
A Mediapart investigation reported that Mistral AI trained its models on books obtained from Library Genesis (LibGen), a pirated-book database, citing internal emails from co-founder Guillaume Lample describing use of the LibGen corpus. French publisher Nouveau Monde Editions said it found more than 200 of its titles - about 20% of its catalog - in the LibGen data, and major French publishers including Gallimard, Seuil, PUF, Les Belles Lettres, Armand Colin and La Decouverte said their works were 'massively pillaged.' Mistral denied any agreement with Nouveau Monde and said its models are trained on 'public sources, licensed content, and internally generated data,' but has not published the training-data summary required under the EU AI Act since August 2025.
On September 4, 2025, Warner Bros. Discovery filed an 87-page copyright infringement lawsuit against Midjourney, claiming the company is engaged in the 'theft' of its intellectual property including Superman, Batman, Wonder Woman, Bugs Bunny, Scooby-Doo, Tom and Jerry, Powerpuff Girls, and Rick and Morty. The complaint alleged Midjourney's infringement was 'systematic, ongoing, and willful.' When Midjourney launched video generation capabilities, the service initially refused to animate many infringing images of Warner Bros. characters, but within weeks allegedly removed these protection measures and announced it as an 'improvement.' Case consolidated with Disney lawsuit on November 4.
reactive $1.5B
In August 2025, Anthropic agreed to pay $1.5 billion to settle the Bartz v. Anthropic class action, the largest copyright settlement in US history. The settlement covered approximately 500,000 copyrighted works at ~$3,000 each. Anthropic also agreed to destroy the two pirated book libraries and derivative copies within 30 days. The settlement only covered past conduct and did not create an ongoing licensing scheme. Judge Alsup granted preliminary approval in September 2025.
negligent
In mid-2025, an AI-generated music project called 'Velvet Sundown' accumulated over 1 million streams on Spotify and received a verified artist badge, despite being entirely created by artificial intelligence without human musicians. The case highlighted Spotify's inadequate systems for detecting and labeling AI-generated content, raising concerns about AI music displacing human artists and misleading listeners about the nature of what they're hearing.
In 2025, multiple major studios filed copyright infringement lawsuits against Midjourney. Disney, NBC Universal, and DreamWorks sued June 2025 alleging 'mass infringement.' Warner Bros. sued September 2025 claiming Midjourney 'willfully creates both still images and video' of Superman, Batman, Bugs Bunny, Daffy Duck, Tom and Jerry. Studios alleged Midjourney 'will readily generate images of copyrighted characters by name'—users need only prompt 'Yoda' not 'short, elderly, green humanoid alien.' Complaint states 'Midjourney has no internal protocols intended to prevent such use of the platform.' Artist class action from January 2023 also ongoing. Midjourney defense asserts fair use, comparing AI training to how humans learn art.
On June 11, 2025, Universal Pictures and The Walt Disney Company filed a copyright infringement lawsuit against Midjourney in U.S. District Court for Central District of California, accusing the company of 'massive and deliberate copyright infringement' for allowing users to create highly realistic depictions of proprietary characters including Elsa, Shrek, Darth Vader, and Homer Simpson without license or authorization. The suit described Midjourney as 'a bottomless pit of plagiarism.' Both studios had served a cease-and-desist order which Midjourney reportedly ignored.
negligent
In June 2025, a Guardian investigation exposed how AI-powered bot farms were systematically uploading thousands of AI-generated tracks to Spotify and using automated streaming bots to generate fraudulent royalty payments. The investigation found organized operations generating millions of fake streams, diverting royalty money from legitimate artists. Despite Spotify's claims of anti-fraud measures, the investigation showed the platform's detection systems were largely ineffective against sophisticated AI fraud operations.
On February 13, 2025 a coalition of major US news publishers including Condé Nast, The Atlantic, Forbes, The Guardian, Los Angeles Times, McClatchy, Politico, The Republican Company, Toronto Star and Vox Media filed a copyright infringement lawsuit against Cohere in the Southern District of New York. The complaint alleged Cohere ingested and reproduced full articles from at least 4,000 publications and that the company's hallucinations falsely attributed fabricated content to named publishers. The suit was among the largest publisher-led actions against an enterprise AI company.
Reddit filed a lawsuit in California state court against Anthropic, alleging the AI company made over 100,000 unauthorized requests to Reddit's servers to collect user posts and comments without permission. The suit alleged Anthropic circumvented Reddit's robots.txt file and refused to engage in licensing negotiations, unlike Google and OpenAI which entered formal licensing agreements. The case raised questions about intellectual property rights and data protection for user-generated content.
negligent
Universal Music Corp., ABKCO Music Inc., Concord Music Group, and other music publishers sued Anthropic in federal court, alleging Claude reproduces copyrighted song lyrics without proper licensing. On December 30, 2024, Anthropic agreed to a partial injunction while continuing to fight the copyright infringement claims. The case is ongoing in the Northern District of California.