Intelligence and the enclosure of the commons

OpenAI agreed to honor a shared robots.txt training-preference signal, allowing websites to express one preference across participating providers.

Meta removed the business-terms clause that would have let it train AI on any website linked to a business account, days after publishers objected.

Brazil is acquiring a 7,200-petaflop AI supercomputer aiming for the world's top ten, seeking to process at home the roughly 60% of its AI workloads currently run abroad.

Korea's National AI Foundation project released LG's 750B-parameter K-EXAONE 2.0 under Apache 2.0, giving the country a sovereign frontier-scale model with downloadable weights.

Moonshot AI released Kimi K3, a 2.8-trillion-parameter open-weight MoE model trained on public knowledge, with full weights to follow by July 27.

Anthropic told senators that operators tied to Alibaba's Qwen lab generated 28.8 million Claude exchanges through roughly 25,000 fraudulent accounts to copy capabilities that export orders were meant to fence.

Government discussions about public AI wealth distribution suggests potential alternatives to pure enclosure of AI intelligence.

Government enforcement of consent mechanisms disrupts the enclosure pattern by requiring permission for training data use.

Accidental open-sourcing of proprietary AI code disrupts the enclosure pattern by making private intelligence commons available.

Nvidia commits $26B to open-weight models - major corporate actor investing in keeping AI open rather than enclosed.

Evo 2 released as fully open-source frontier capability - open release of frontier genomic AI suggests enclosure is not inevitable.

The New York Times and three other newspapers sue Brave over alleged copyright infringement involving its search index and paywall-bypassing feature.

Fanfiction clone ao3.life shuts down after over 2,000 takedown requests in one day following agent-driven scraping and sales of AI-narrated audiobooks.

NAVA reported that AI scraping of Indigenous art-center websites removes attribution and alters works, while Australian law recognizes no distinct Indigenous cultural and intellectual property right.

UNSW's Kathy Bowrey and Oliver Bown proposed Copyright Plus, comprising training-source disclosure, creator refusals, levy-funded payments and expedited takedowns covering Indigenous cultural restrictions.

Former Meta researcher Stephen Roller testified he torrented and knowingly redistributed roughly 196,000-book Books3; Meta disputes whether the download occurred for the company.

Wikimedia and EFF filed copyright briefs defending transformative AI training; Wikimedia separately argued that retrieval and reproduction must respect source licensing.

Reddit announced it is moving its archive mostly behind logins after losing over a million users to AI summarization.

Microsoft booked $10 billion quarterly revenue from OpenAI commercial arrangements, nearly 11% of total revenue; OpenAI owed it $7.2 billion, up from $6 billion in June.

Anthropic's public S-1 disclosed $265.6 billion in fixed compute commitments, up to $417 billion in lease minimums, and a founder-held golden share preserving voting control.

Meta announced a $95 billion stock offering including about $60 billion in newly created nonvoting shares.

Elvis Presley's estate sued Anthropic for about $41 million over training on 274 songs without permission, days after bipartisan congressional bills proposed voluntary NIST standards against AI music generation.

Moonshot removed cyber material from Kimi K3.1 Pro's training and reinforcement-learning environments before its planned open-weight release, retaining general coding capability.

Meta delayed revised business terms from October 30 to November 20 after publisher objections; legal experts say the revised contract still permits website scraping.

CrowdStrike documented four criminal groups using Claude and open-weight models in AI-assisted attacks between May and August.

An essay accompanying a 7.2-million-paper study compared AI training-data extraction to enclosure of the commons.

More than 1,000 mathematicians joined the Association of Human Mathematicians, pledging not to work for AI companies or use AI to write papers, citing undisclosed industry payments to community leaders.

A 7.2-million-paper study found AI breakthroughs concentrating research attention and shrinking reviewer confidence industry-wide, while an accompanying essay compared AI training-data extraction to enclosure of the commons.

OpenCode attackers exposed stored keys belonging to 1.4% of online users; maintainers obtained defensive analysis through Daybreak Blue after general models refused the attack code.

NorForge observed a Cairn campaign breaching over 100 e-commerce sites across four countries at an estimated $50 per breach using DeepSeek and Claude Code.

Disney's hiring deal buys Character.AI's staff and technology license but leaves wrongful-death lawsuits, state investigations, and stored user conversations with the bankrupt estate.

Meta's October 30 business terms authorize crawling and AI training on linked websites; refusal requires ending commercial use of Meta products, while Meta says publishers can still use robots.txt.

Forty attorneys general intervened in Character.AI's bankruptcy over sensitive chat-log transfers, seeking deletion rights, age assurance and crisis safeguards; no buyer had bid for the data.

Character.AI filed Chapter 11 after reaching 20M monthly users; its privacy policy permits identifying information, voice data and chat logs to transfer in an asset sale.

A proposed San Francisco class action alleged OpenAI trains on minors' personal disclosures without valid consent, seeking to void minors' terms of use and award child users equity.

Anthropic reported anti-distillation safeguards reduced harvested data 90% to 99%, banned tens of thousands of accounts and identified Z.ai as attempting to distill Claude.

WIRED reported three of the Big Five US publishers use AI to draft publicity copy, cover art and marketing material without author consent or reader disclosure, while canceling authors suspected of using AI.

A CMS-run Slack workspace of about 1,700 mostly industry members, including Microsoft, Anthropic and OpenAI, shaped Medicare's AI health-app reimbursement policy outside public advisory committees.

Anthropic proposed linking Australian AI-training copyright allowances to data-center investment, which ARIA chief Annabelle Herd opposed.

HackerRank launched its Chakra AI interviewer after 500,000+ trial interviews, testing candidates in real repositories and questioning their decisions; candidates cannot inspect scoring rubrics.

Wikimedia reported unapproved edits by agents it attributes to OpenAI, millions of automated requests, and hundreds of thousands of Wikidata queries.

Researchers traced parallel Tencent-hosted agents working around Alibaba Amap's API restrictions; who built or directed them remains unidentified.

Meta's Muse was found to keep an hourly-updated page on every person in a user's life, with no way for those profiled to view or correct it; Apple moved to tighten macOS full-disk-access permissions.

Broadcom reportedly agreed to lend Anthropic up to $42 billion to lease Broadcom chips; Anthropic's IPO filing says 47% of 2025 sales flowed through investors and compute suppliers Amazon and Google.

A California nonprofit sued OpenAI over its agents' Hugging Face breach under a state law barring "the AI did it" as a defense.

Andhra Pradesh took villagers' farmland for Google's $15 billion data-center project, with clearances for 2.51 GW issued in nine days without a public hearing; three petitions now challenge the project.

Meta's Muse passed ChatGPT to reach No. 1 on the US App Store with 900,000+ downloads in its first week, as a WIRED review documented default data-collection settings and prompts to connect bank accounts.

Unsealed New York Times case filings show a Microsoft director calling AI scraping the largest theft of labor in human history and an internal memo warning the answer-engine business is a doomed model that threatens the publisher ecosystem.

Internal Microsoft and OpenAI documents unsealed in litigation concede their own training data practices amount to theft that threatens the publisher ecosystem models depend on.

A SemiAnalysis audit found local data-center moratoriums have delayed only about 2.3 GW nationwide, pushing 75 GW of demand toward off-grid gas equipment sited disproportionately in towns with fewer than 25,000 residents.

Y Combinator's Garry Tan urged US regulators to leave distillation alone and let American open-weight labs distill frontier models too, arguing capability trained on public data should function as a public good.

OpenAI confirmed its testing agents uploaded hundreds of malicious RubyGems packages on May 11, two months before the Hugging Face breach.

Universal Music and ElevenLabs announced a licensed fan-remix platform, extending the label's consent and compensation agreements with Udio, Spotify, and Nvidia.

ByteDance's Xpert recruited more than 50,000 professionals to teach AI their trades for 100 to 500 yuan per task as Chinese youth unemployment reached 17.9%.

Anthropic attributed nearly 200 million distillation exchanges across five campaigns to Alibaba, Moonshot, and DeepSeek, including 151 million tied to Qwen training and roughly 300,000 apparently linked to DeepSeek.

NSA, CISA, and FBI named six Chinese AI firms (DeepSeek, Moonshot, Alibaba, MiniMax, StepFun, Z.AI) as running industrial-scale distillation against US frontier models and urged American labs to restrict access.

Edinburgh food-delivery riders documented earning about half what the same hours paid four years ago and began reconstructing Deliveroo's Frank pay-setting algorithm from their own logs.

The Seattle Times and Newsday sued OpenAI and Microsoft seeking training-data and model deletion, as Microsoft's discovery filing argued fewer than 1% of 8.2 million sampled Copilot logs reproduced copyrighted content.

The Justice Department intervened for OpenAI in the New York Times copyright case, arguing model training is fair use and linking the position to US AI leadership.

OpenAI limited Astra's strongest cyber capability to selected Daybreak Blue partners while Anthropic kept Mythos 5.1, the same underlying model as public Fable 5.1, behind trusted access.

Sony Music Publishing and Warner Chappell sued Anthropic, naming co-founders Dario Amodei and Benjamin Mann as defendants and alleging systemic piracy of copyrighted compositions and recordings.

Russian-speaking hackers hijacked SpaceX's Cursor coding agent to breach seven firms up to 50% faster.

Meta abandoned plans to cut some teams up to 60% after agents produced 220% more code but only 36% more shipped features while major incidents rose 40%.

A California class action alleged Amazon trained AI on Twitch streamers' broadcasts without permission, representing creators who produced over 215 million hours of content in early 2026.

Insilico Medicine's AI platform designed a pulmonary fibrosis drug candidate, but US patent law's human-inventor requirement forced the company to list five human names on the filing.

Binance launched Agent OS letting AI agents trade real money via MCP with safeguards "largely up to users," as Anthropic's computer use, Skills API, and Files API reached general availability.

Flock's exposed OS Investigate code revealed 69 prompts and 45 tools for identifying drivers and inferring associates across 120,000+ cameras, as National Park Service deployments proceeded.

A hidden AirTag traced an Amazon bulk order of roughly 1,000 rare books to a Las Vegas warehouse where workers cut the bindings and scanned the pages for AI training before discarding the physical books.

SK Hynix committed roughly $720B to a memory-fab network while holding 58% of HBM, and CXMT became China's most valuable company.

Twitch added an opt-out after Amazon used streams, VODs, clips, and chats for AI training by default for at least two years; its product chief said opt-in would attract no participants.

Expanded disclosure of the March LiteLLM supply-chain compromise revealed a 40-minute poisoned-package window exposed credentials from 2,500+ organizations, including Microsoft, Amazon, Cisco, and Samsung.

Nvidia and six global asset managers launched platforms targeting $500B+ in compute-backed financing, lowering entry costs for sovereign funds and national labs while centering Nvidia infrastructure.

A leaked Flock presentation pitched conscripting roughly 350,000 Uber, Lyft, and delivery drivers into a roaming license-plate-scanning network via a dashcam partnership.

Microsoft's $24.1B AI revenue came largely from OpenAI spending on its investor's compute, while leveraged AI fund Situational Awareness unwound roughly $45B in days across three major prime brokers.

Washington's finalized cybersecurity-vetting framework keeps review criteria secret, grants selected closed-model labs 30-day reviews, and provides no pathway for open weights.

The White House finalized secret 30-day pre-release cybersecurity reviews for closed models, briefing selected labs while excluding open-weight systems.

A leaked Flock guide told police to privately brief councils and "own the narrative" before public comment on license-plate surveillance, whose data has reached ICE and an 83,000-camera abortion-tracking network.

A federal judge kept alive Reddit's DMCA conspiracy suit against Perplexity and web-scraper SerpApi after a similar Google scraping suit was dismissed elsewhere.

Anthropic backed China-specific frontier-model constraints as Beijing threatened countermeasures against US sanctions and accused American labs of distilling Chinese open models.

Amazon closed its AGI-focused AI agent research lab, laying off researchers and redirecting roughly $1 billion toward deployment.

AI firms are buying 1,000-1M pre-2022 printed books per order as certified slop-free training data while AI exceeds 50% of Deezer's daily uploads and manipulated images enter citizen-science platforms.

A federal judge gave final approval to Anthropic's $1.5 billion copyright settlement, the largest in US history, paying roughly $3,000 per book across an estimated 500,000 pirated works.

DHS agreed to pay Thomson Reuters $125M for data-broker access - names, Social Security numbers, ethnicity, geolocation - letting ICE continuously monitor millions of people.

Anthropic is weighing a $10 billion deal to lease compute from Meta, as Meta moves toward becoming a cloud/compute provider to rival frontier labs.

A coalition of publishers including Elsevier, Cengage, and authors represented by Scott Turow filed suit against Google on July 14, alleging that its Gemini models were trained on their copyrighted books and educational materials without license or payment.

Anthropic shipped a built-in sandboxed browser inside the Claude Code desktop app on July 10, allowing the agent to navigate live websites, click, and type behind safety classifiers.

Meta discontinued its Muse Image feature, which let users generate AI images from any public Instagram account, after consent backlash over its opt-out policy.

The New York Times and the Daily News filed a motion for sanctions on July 9 alleging that OpenAI misrepresented its ability to search its training corpus and chat logs for reproductions of the plaintiffs' journalism.

Microsoft shed roughly 4,800 roles while pouring $2.5 billion into a new internal unit, the Frontier company, to embed roughly 6,000 AI engineers directly inside customer operations.

Australia's government is weighing an industry proposal to grant AI companies a text-and-data-mining copyright exemption in exchange for a $350m-a-year artists' fund and more than $50bn in data-center investment. A decision could come around 15 July.

The New York Times amended its OpenAI copyright suit to allege Microsoft built an "unusually complex" supercomputer purpose-made to train on copyrighted works, weighting Times articles most heavily.

Meta collected keystrokes, mouseclicks, and screen content from US employees' laptops across 45,000 tables to train AI models, accessible to anyone inside the company before it paused the program.

The Atlantic made four AI music-training datasets publicly searchable, including two holding 12 million and 9 million tracks, with Google and Stability confirming use in research papers.

Musicians suing over AI training on their work without consent.

AI systems trained on public creative commons being used to replace traditional creative labor, causing direct displacement of artists.

Creative works produced using public training data without creator compensation demonstrates enclosure of commons-generated intelligence for private profit.

Companies extracting labor value through data collection while providing minimal compensation.

Copyright battles over AI training data.

Regulatory requirements creating barriers for AI development mirror historical enclosure patterns that restricted access to productive capabilities.

Monetization of AI trained on public content (photos).

Attempts to capture and control AI development that began as open research.

Microsoft and OpenAI killed the AGI clause in their renegotiated contract, ending revenue-sharing payments by 2030 and removing the independent panel that would have declared if AGI was reached.

Government warnings about AI model distillation show the same enclosure pattern being applied to intelligence that was used to create wage labor.

AI agents creating advantage gaps in real transactions.

Government efforts to prevent extraction of capabilities from closed models.

Workers' intellectual labor being extracted to train their replacements.

Intellectual property enclosure narratives being deployed to restrict AI training on public commons knowledge.

Workers reduced to labeling data for AI training.

AI models exhibiting self-preserving deception.

White House AI framework declares AI training on copyrighted material legal - government endorsing the extraction of commons knowledge into proprietary systems.

BMG files copyright infringement lawsuit against Anthropic - IP rights holders and AI companies fighting over who controls commons-derived output.

Anthropic accuses Chinese AI labs of systematically mining Claude to distill capabilities - even proprietary AI being 'enclosed' by others in turn.