Organized by Erik Brynjolfsson, Ajay Agrawal, Anton Korinek, and Tom Cunningham, the statement warns AI could reshape the economy faster than the Industrial Revolution and urges economists, policymakers, and tech leaders to prepare now. Signatories include Nobel laureates Michael Spence and Daron Acemoglu; it was echoed July 14 in Fortune under the headline “Almost 200 economists warn of AI-driven job displacement.”
Snapshot — July 13, 2026
94 stories
- The AP reports that ~$700B in 2026 data‑center investment — led by Alphabet, Amazon, Meta and Microsoft (~$720B combined) — has pushed up prices for memory, processors and electricity, with JPMorgan estimating some memory‑chip costs could rise as much as 400% by year‑end.
- Apple has already raised MacBook and iPad prices 15–25%, and Microsoft is lifting Xbox prices, both citing memory costs.
- The New York Times reports on what the government's fight with Anthropic reveals about free speech in America.
- The article points to a growing policy question: how far governments can go in pressuring, regulating, or conditioning AI model behavior without crossing speech and viewpoint boundaries.
- For AI companies and enterprise buyers, this signals that legal risk around model outputs is expanding beyond copyright and safety into constitutional and governance debates.
- Singapore-based PixVerse said it closed a Series C extension totaling $439 million, pushing its valuation past $2 billion on the strength of roughly 15 million monthly active users.
- New investors include Alibaba, and the company plans to expand its “world model” offering and reach customers across more geographies.
Airlines and Hyperscalers Clamor for Turbines. Only a Few Companies Make the Parts.
Alibaba posted its largest stock jump since September after an earnings update, buoying Baidu, Tencent, and JD.com as investors rotate toward cheaper Chinese AI. Qwen is now cited as the world’s most-downloaded AI model, aided by aggressive marketing and per-token costs far below US rivals — even as a US House committee probes American firms’ use of Chinese models.
- Anthropic has started showing local pricing for Claude in India — its largest market outside the U.S. — on its website and mobile apps.
- Notably, it has not yet enabled UPI, India’s dominant instant-payment rail, unlike OpenAI, which rolled out rupee pricing with UPI support last year.
- The move signals intensifying competition to convert usage into paid subscribers in the world’s most populous market.
Anthropic extended free subscriber access to Claude Fable 5 for the second time in a week — an apparent response to OpenAI’s public release of its GPT-5.6 “Sol” flagship, which OpenAI claims sets a “new standard.” Fable’s debut had been overshadowed by a clash with the U.S. government, which briefly blocked Anthropic’s restricted “Mythos” models. The tit-for-tat underscores how quickly the two leaders are now trading blows for developer mindshare.
- Anthropic gave Pro, Max, and Team subscribers a second one-week extension of free access to its flagship "Mythos-class" Claude Fable 5 model, moving the cutoff from July 12 to July 19, alongside a 50% boost to Claude Code weekly rate limits.
- It is the second extension in a week, widely read as a competitive response to OpenAI's GPT‑5.6 launch.
In a new report on behavioral inconsistencies published Monday, Anthropic said Claude’s expressed values differ by the language a user writes in — for example, responding with more warmth in Hindi and more rigor in Russian — mapping hundreds of value concepts onto four core dimensions. Researchers cautioned that these “imbalances” may run deeper than tone and could shift the model’s priorities, adding they “aren’t yet sure how much of this variation is desirable.” For global deployments, the finding underscores that multilingual rollout is a governance question, not just a translation one.
- TechCrunch summarized unusually detailed allegations in Apple's trade-secret lawsuit against OpenAI, including claims about former Apple employees, access to internal systems, interview requests involving Apple hardware or design artifacts, and OpenAI's acquisition of io.
- OpenAI has said it has no interest in other companies' trade secrets.
- Apple's complaint alleges a former employee exploited a rare bug to download confidential files after leaving for OpenAI, tied to Apple's unreleased AI-hardware program.
- The case could set a template for how courts view senior technical hiring, device ambitions, and trade-secret controls in the AI talent wars.
- arXiv's cs.AI section shows 116 new preprints dated Monday, July 13, concentrated on agentic LLMs, reasoning, and trustworthy/safety-oriented AI.
- In-window examples include "Multimodal Reward Hacking in Reinforcement Learning," "ProofCouncil: An LLM Agent for Solving Open Mathematical Problems," and "Toward Auditable AI Scientists: A Hypothesis Evolution Protocol for LLM Agents." These are unreviewed preprints, noted as a research pulse rather than validated breakthroughs.
Amazon launched a UI in SageMaker AI Studio that walks teams through preset use-case profiles, visual benchmark comparisons, and one-click deployment for optimized inference — removing the need to hand-tune parameters. A smaller but firmly in-window enterprise tooling update.
Berkeley RDI - [2026-07-13] *(No new Berkeley RDI emails found for 2026-07-13)*
Business Insider - [2026-07-13] [EXTERNAL] Today: Apple's explosive allegations - [2026-07-13] [EXTERNAL] The four side hustles worth trying
- Meta will scale Hyperion to 5 GW at >$50B — up from initial ~$10B/2 GW.
- Partner Entergy will build ~7 GW of new generation.
- Aggregate Louisiana AI commitments now exceed $250B.
- The physical layer, not model architecture, remains the binding constraint.
CIO Dive - [2026-07-13] [EXTERNAL] July 13 - Agentic AI strains legacy IT | Dev teams will shrink
Coalition of groups representing music labels and artists says fans want transparency.
More than 200 researchers and economists, including 15 Nobel laureates and leaders from OpenAI, Anthropic, and Google DeepMind, issued a joint statement urging governments and technology leaders to address AI's economic effects. They warned that AI could drive a transformation larger than the Industrial Revolution but on a much shorter timeline.
*Coverage from newsletter subscriptions for 2026-07-13*
Cursor’s internally code-named “Sand” is a general-purpose assistant for emails, texts, spreadsheets, and documents aimed at non-developers — its first product beyond coding. Internal rollout began in late June; a public launch remains uncertain given the pending ~$60B acquisition of Cursor's parent, Anysphere.
- Governance and distribution move to center stage as the model race cools.
- Hassabis proposed a FINRA-style oversight body for frontier AI;
- Nadella warned enterprises that proprietary model vendors carry a "Trojan horse" risk.
- Capital kept flowing to applied AI (Chai Discovery $400M, PixVerse $439M, Nous Research $1.5B), while Apple–OpenAI sharpened.
DealBook (Andrew Ross Sorkin / NYT) - [2026-07-13] [EXTERNAL] DealBook: Wall Street's big week
- DeepSeek is reportedly weighing another raise at a ~$71B pre-money valuation;
- Chai Discovery raised $400M at $3.8B for AI drug design;
- PixVerse raised $439M at a $2B+ valuation; and Nous Research is reportedly raising at a $1.5B valuation.
- Capital is spreading from frontier models into applied vertical AI, video, open agents, and China-based model ecosystems.
DeepSeek is reportedly in preliminary talks to raise at roughly a $71 billion valuation — about a $19B markup from the ~$52B post-money set in late May, when it closed its first external round (~$7B, led by Tencent and CATL). Separately, Bloomberg reported July 14 that founder Liang Wenfeng has overtaken Dario Amodei and Greg Brockman as the richest AI founder.
- More than 200 economists and AI researchers, including 16 Nobel laureates and leaders from OpenAI, Anthropic, and Google DeepMind, signed a statement urging faster preparation for AI's economic impact.
- The breadth of signatories reframes AI labor disruption from a research topic into an active governance demand.
EU regulators forced Meta to restore ChatGPT access within WhatsApp after Meta had blocked it. The competitive battle is shifting from raw model capability to distribution, platform access, and interoperability.
German AI-defense startup Helsing said it raised $1.8 billion at an $18 billion valuation, with new and existing backers including JPMorgan Chase, Lightspeed Venture Partners, and Iconiq. The company said "investor demand significantly exceeded the available allocation, reflecting strong and growing confidence in AI-driven and software-defined defence technology," cementing its position as Europe's largest defense startup and a direct rival to Anduril.
- Researchers coupled a generative adversarial network to latent vectors sampled from a real 32-mode photonic quantum processor (boson sampling) to design MHC class I-binding peptides.
- Across 131 HLA alleles in silico, the quantum-derived priors increased the yield of predicted strong binders — with the largest gains for understudied alleles.
- Soofi S 30B-A3B activates 3.2B of 31.6B parameters per token and tops fully open models on German and English benchmarks.
- Trained on Deutsche Telekom's Munich cloud using ~512 Nvidia B200 GPUs with a hybrid Mamba-Transformer architecture claiming ~8× throughput vs comparable dense models.
- A deliberate European sovereignty play.
- Google expanded Gemini in Chrome to U.K. desktop users, added Gemini-powered Waze search and route updates, and emphasized data-localized enterprise AI in India.
- The thread is distribution: Google is embedding Gemini into high-frequency workflows where defaults and context matter more than standalone benchmarks.
- The Information reports that Google is mounting a TPU campaign to win customers historically committed to Nvidia GPUs.
- The competitive importance is not just chip substitution; it is a broader attempt to use vertically integrated cloud infrastructure to reshape AI compute purchasing.
- If successful, the effort could increase buyer leverage and pressure Nvidia's software-and-ecosystem moat.
- Google DeepMind is reportedly targeting July 17 for general availability of Gemini 3.5 Pro after a full rebuild, but Tech Times cautions that every circulating detail — the date itself, a rumored 2-million-token context window, and benchmark figures — comes from third-party reporting, not an official Google announcement.
- Waze introduced an AI‑built Motorcycle mode that accounts for rider‑specific shortcuts and hazards (potholes, speed bumps, narrow bridges), a "Less Chatty" voice option, and route suggestions based on prior trips — part of a broader push to embed Gemini across Google's navigation apps.
- The features still lean on Waze's real‑time traffic map and human, motorcycle‑focused editors for accuracy.
WSJ highlighted surging demand for specialized turbine parts as airlines and hyperscalers compete for a limited supplier base. The investable bottleneck is widening beyond GPUs, memory, and packaging into the industrial supply chain that supports power generation for data centers.
- Axios reports that New Orleans has paused data-center activity amid rising local concern over AI-driven infrastructure demand.
- The story fits a broader pattern: local zoning, water and power availability, and ratepayer exposure are becoming practical choke points for national AI buildout.
- For enterprises depending on cloud AI capacity, permitting friction is now part of infrastructure risk.
- WSJ reports that Meta has raised the expected cost of its Louisiana data-center project to $50 billion, highlighting how quickly AI infrastructure commitments are scaling.
- The reported figure reinforces the capital intensity of frontier AI deployment and the growing dependence on state incentives, power access, and local permitting.
- Intel began a €5 billion capital program at its Leixlip campus to expand Intel 3 output for Intel Xeon 6 and next‑generation Xeon processors, citing surging demand for silicon to power "AI Factories." The spend is about 30% of Intel's ~$17B planned 2026 capex, with most deployed by end‑2027, and adds several hundred jobs to the 4,900 already on site.
- Intel announced €5B to expand Fab 34 for Xeon 6 and next-gen processors on Intel 3 node.
- Follows April's $14.2B Apollo stake buyback.
- Total Ireland investment >€30B since 1989 — a foundry-strategy statement as much as a capacity one.
- Meituan unveiled LongCat-2.0, which it calls the industry’s first trillion-parameter model to complete its full training and inference lifecycle on a 50,000-card domestic computing cluster.
- It carries 1.6T total parameters (33B–56B dynamically activated), natively supports a 1M-token context window, and is architected specifically for “agentic coding” tasks.
VitaBench 2.0 is billed as the first benchmark focused on real-life, evolving user interactions rather than static one-shot tasks, evaluating agents on personalization and proactivity over extended periods. It shifts agent evaluation away from short-term task completion toward sustained human–AI engagement — a metric that matters for consumer and assistant products.
Memory-chip companies are expected to post large sales gains as earnings season begins. The continued pull-through from AI servers into high-bandwidth memory supports the view that AI infrastructure spending is broadening beyond GPUs into the full compute stack.
- Meta will scale Hyperion to 5 GW at >$50B — up from initial ~$10B/2 GW.
- Partner Entergy will build ~7 GW of new generation.
- Aggregate Louisiana AI commitments now exceed $250B.
- The physical layer, not model architecture, remains the binding constraint.
- A Meta patent application (US 2026/0182881, published July 2) describes an AI that reads emotional state from a user's voice — each read timestamped and tagged with location, activity and phone use — with some variants listening throughout the day.
- Meta has not announced a product, and a filing only stakes a claim on an idea, but the disclosure sharpens privacy and consent questions around ambient AI and wearables.
Meta is adding roughly $40 billion to its Hyperion campus in Richland Parish, taking total site investment from about $10B toward as much as $50B and lifting aggregate regional commitments past $250B. The 3,200-acre site is more than four times the size of Central Park — a vivid marker of the hyperscaler compute arms race.
Internal documents show Meta plans to begin manufacturing its custom data-center accelerator, codenamed Iris, in September as part of a four-generation MTIA roadmap scaling toward 14 GW of compute by 2027. Built with Broadcom and TSMC, it reportedly passed testing in six weeks — Meta’s most aggressive push yet to reduce reliance on Nvidia and AMD GPUs.
- Forbes examined an apparent contradiction in Meta’s recent moves: the company committed more than $50 billion to expand a Louisiana data center to 5 gigawatts while simultaneously revealing plans to sell “excess” AI compute.
- The analysis frames this as a bellwether for a broader question — how much acquired AI compute is actually being used — with the absence of clear utilization metrics making it hard to separate strategic foresight from overbuilding.
- Business Insider reports that Microsoft CEO Satya Nadella criticized, indirectly, AI model makers whose value is concentrated in foundation models rather than products, distribution, and workflow integration.
- The comments matter because Microsoft is positioning enterprise AI advantage around application surfaces, cloud infrastructure, and customer workflows, not only model access.
Mistral released an 8B embodied-navigation model that uses a single RGB camera and natural-language instructions to move robots through real environments. The company says it reaches 76.6% success on R2R-CE validation unseen and was trained entirely in simulation across roughly 400,000 trajectories.
- Researchers at MIT CSAIL and the Toyota Research Institute introduced SceneSmith, which orchestrates three vision-language-model agents — a “designer,” a “critic,” and an “orchestrator” — to auto-generate realistic, object-dense 3D indoor scenes for robot simulation.
- The team produced more than 1,300 scenes with up to six times more objects than prior methods, and pretrained robot policies operated in them without prior exposure.
- MIT News profiled the MIT Cybersecurity Clinic, which provides free vulnerability assessments to municipalities, hospitals, and other resource-constrained organizations.
- The article notes that AI is changing both cyber offense and defense, with AI increasingly able to identify vulnerabilities and assist attacks while organizations still struggle with human-centered security practices.
- MIT researchers introduced an auditing method that detects whether a generative model has been adapted to produce CSAM without generating illegal outputs.
- It analyzes LoRA adaptor behavior through hidden representations and reportedly identified specialized variants with 100% accuracy.
- Model Releases NEW ROBOTICS
- MIT researchers, working with child-safety nonprofit Thorn, developed an auditing technique that determines whether an open-source generative model has been fine-tuned to produce child sexual abuse material — without ever prompting it to generate illegal output.
- The method uses "Gaussian probing" of a model's LoRA adaptor modifications and hidden representations, and identified maliciously specialized variants with 100% accuracy in testing.
- Today's cycle is about the economics of AI rather than new model launches.
- TSMC posted record quarterly revenue and Intel committed €5B to expand European fab capacity, even as the Associated Press flagged that roughly $700B in 2026 data‑center spend has become a measurable inflation risk feeding into the Fed's rate path.
Microsoft CEO Satya Nadella warned enterprises that proprietary models can deepen a lab's access to sensitive data where the model maker could become a competitor. It positions trust, control, and workflow ownership as strategic differentiators.
- Satya Nadella warned that companies using proprietary AI models may pay once in token fees and again by exposing proprietary knowledge through prompts, corrections, tool use, and workflow data.
- Microsoft is framing enterprise AI advantage around control of data, routing, and proprietary learning loops.
MIT researchers developed a technique to test open-source generative models for dangerous or illegal capabilities — such as producing hate speech or CSAM — without ever prompting the model to actually output that content. As open-weight models proliferate and can be cheaply fine-tuned by bad actors, it gives model hosts, platforms, and regulators a safer, legally defensible way to screen a model before release, turning a fast-growing trust-and-safety risk into something that can be measured proactively.
- 200+ economists and AI researchers, including 16 Nobel laureates and leaders from OpenAI, Anthropic, and Google DeepMind, issued a joint statement urging faster preparation for AI’s economic impact.
- They argue AI may transform labor markets faster than prior general-purpose technologies.
- The breadth of signatories reframes labor displacement from a research topic into an active policy demand.
Nous Research, maker of the Hermes agent, is reportedly finalizing at least $75 million at a $1.5 billion valuation. It sits at the intersection of open models and agentic infrastructure, a category investors are pricing at unicorn scale.
Oil jumps as Trump vows to renew Iran blockade - Nasdaq slumps as AI doubts spoil SK Hynix's big U.S. debut - Persian Gulf strikes rattle markets
The claimed proof of the 50-year-old cycle double cover conjecture by GPT-5.6 Sol Ultra, using 64 coordinated subagents, continues drawing attention and skepticism. Whether or not it holds, multi-agent orchestrated inference is now a serious academic topic.
Other AI-related Publication Emails - [2026-07-13] *(No other clear AI-related publication emails found for 2026-07-13)*
PitchBook - [2026-07-13] [EXTERNAL] AI's musical chairs
Verifiers v1 provides composable tasksets, harnesses, and runtimes for agentic reinforcement-learning training and evaluation, including task management and training-trace recording. It aims to standardize how developers build and benchmark interactive agent environments.
Princeton merged five units — the Survey Research Center, the Center for Statistics and Machine Learning, the Data-Driven Social Science Initiative, PICSciE, and the AI Lab — into a new academic unit, Data and Intelligent Systems (DaIS), split into statistics/data-science and AI divisions. It is a concrete data point on how elite research universities are centralizing AI strategy into formal org structures, though reporting flags limited faculty consultation and unclear governance. (Sourced from the student newspaper, not an official Princeton research announcement.)
- Monzo/GoCardless co-founder Tom Blomfield leaves YC to join Anthropic's compute team as "member of technical staff," echoing Karpathy and ex-Workday CTO Bailis.
- The flat title signals how strongly the frontier is pulling senior operators into IC work.
- A read on where scarce AI talent is concentrating.
China's foreign ministry said President Xi Jinping will attend the opening ceremony and deliver a keynote at the 2026 World AI Conference and High-Level Meeting on Global AI Governance. Analysts expect Xi to advance a Chinese-led World AI Cooperation Organization, positioning open weights and multilateral membership against Washington's export-control regime.
A coalition representing music labels and artists is pushing streaming platforms to label AI-generated songs, arguing fans want transparency. The effort could establish expectations for AI-content disclosure beyond music.
- Samsung will start operations 1–2 years ahead of schedule at the first of six Yongin fabs.
- Part of Korea's ~$880B national chip commitment.
- Land, power, and water — not tooling — are the pacing constraints.
In an X essay that drew roughly 10M views, Microsoft’s CEO argued enterprises using AI “pay twice” — once in cash, and once in the proprietary know-how models absorb from prompts, tool use, and corrections — and laid out a “five Cs” framework. It is a notable strategic framing of the data-leverage question every large AI adopter now faces.
- The Information reports that Samsung is moving up plant capacity plans to 2029, amid continued pressure to expand advanced chip supply.
- While details remain behind the publication's paywall, the timing points to persistent demand from AI infrastructure customers and competitive pressure from SK Hynix and other memory suppliers.
SK Hynix shares fell 12–15% in Seoul, dragging the Kospi into a steep selloff and triggering a circuit breaker, as investors “sold the news” after the AI-memory maker's record ~$26B Nasdaq debut and amid concerns that its HBM4 shipment ramp is lagging Q2 expectations. Rising volatility across the AI memory complex.
MORPHEUS runs “worlds that never reset,” introducing structured non-stationarity and a six-metric evaluation protocol for continual RL. Notably, standard methods (PPO, HER, EWC, LCM) all remained far below the theoretical upper bound — a pointed argument that continual reinforcement learning is genuinely unsolved for real enterprise agents.
*Source: WSJ Markets PM — access@interactive.wsj.com* *AI-relevant: SK Hynix Nasdaq selloff tied to AI memory demand concerns.*
Specialized manufacturers in the sector are a new pick-and-shovel play for investors.
Stanford researchers unveiled TRACE, which turns an agent’s recurring failures into synthetic reinforcement-learning environments, then trains directly against the specific capability gaps causing those failures. The approach is aimed at making agentic LLMs more reliable on long-horizon, multi-step tasks — a core blocker for enterprise agent deployment.
- Around 200 demonstrators marched between the San Francisco offices of OpenAI, Anthropic, and Google DeepMind, calling for a pause on training new frontier models while keeping existing systems available.
- Organized by the group “Stop the AI Race,” the protest broadened beyond safety to encompass job losses, energy and environmental impact, and rising housing costs.
# Subject: Daily AI News Digest – July 13, 2026
Tech investors are reassessing AI investments despite positive long-term outlook (ING). - Investors may see slowdown in EPS growth and decline in valuation multiples as infrastructure spending leads to higher depreciation costs and reduced share buybacks. - Free cash flow expected to be lower as positive effects on revenue growth take time to materialize.
The Information - [2026-07-13] [EXTERNAL] Wall Street's Biggest Move to Blockchain Begins—With Limits - [2026-07-13] [EXTERNAL] Vinod Khosla and Family Agree to Buy Seattle Seahawks for $9.6 Billion
The Tactical Allocation Letter - [2026-07-13] *(No new The Tactical Allocation Letter emails found for 2026-07-13)*
Items materially outside the 24-hour window were excluded, including the FLI AI Safety Index (Jul 7) and Meituan LongCat-2.0 (~Jul 8). Every item links to an original publication.
This file catalogs email subjects received. Full article extraction requires individual email processing via merge_publications.py.*
- TSMC reported second‑quarter revenue of NT$1.27 trillion (~$39.6B), up 36% year‑over‑year and above its own guidance, with June sales alone up 67.9%.
- Analysts at SemiAnalysis note the foundry is "sold out" on its N3 node — targeted by essentially every leading AI GPU and CPU this year — and estimate TSMC is tracking to over $40B in AI‑chip revenue in 2026.
- Microsoft’s July Visual Studio update introduced real-time GitHub Copilot usage tracking and configurable quota alerts, responding to developer complaints after Copilot’s June 1 shift to token-based, usage-based billing; it also added MCP-server trust validation to protect against unexpected tool changes.
- The Information reports that Wall Street is finding more ways to fund AI, including structures such as CLOs and ATMs.
- The signal is that AI infrastructure financing is broadening from direct hyperscaler capex and venture rounds into more complex capital-market instruments.
- That may increase available capital, but it also introduces refinancing, utilization, and counterparty risks if AI demand or pricing assumptions weaken.
- Wall Street Journal / WSJ - [2026-07-13] [EXTERNAL] WSJ Markets Alert: Trump Says He Is Reimposing the U.S.
- Blockade of Iran - [2026-07-13] [EXTERNAL] WSJ Politics: Lindsey Graham Was More Than a Senator.
- He Was a Bridge. - [2026-07-13] [EXTERNAL] 🍿 Markets A.M.: Media Deals: Great for Bankers, Usually Not for You - [2026-07-13] [EXTERNAL] The 10-Point: The Scramble to Find Lindsey Graham's Successor
- Reuters reports the White House is preparing to rally utilities and data-center companies around commitments tied to AI power costs.
- The policy focus reflects growing federal concern that AI load growth could feed through to consumer electricity prices.
- Energy procurement, grid interconnection, and public cost allocation are becoming strategic constraints.
Williams (NYSE: WMB) announced that a Blackstone Credit & Insurance-led group (with Apollo and KKR) will invest $5.34B for a 49% stake in five behind-the-meter gas “Power Innovation” plants being built to feed AI data centers. It is another marker of the energy-financing scramble behind AI compute.
WSJ Pro CyberSecurity - [2026-07-13] [EXTERNAL] AI vs. AI: Hackers Rebrand to Thwart Cyber Defenses
WSJ Wealth Advisor - [2026-07-13] [EXTERNAL] WSJ Wealth Adviser Briefing: Demand for Turbines, Labeling AI Songs, Rotting-Flesh Plant
- Beijing confirmed Xi will open WAIC (July 17–20) and deliver a keynote — his first in‑person appearance since the event began in 2018 — signaling AI's elevation to top‑level statecraft.
- Analysts expect a push to define a China‑led World AI Cooperation Organization headquartered in Shanghai, pitching open‑weight, low‑cost models and "membership" governance to the Global South.
- Zhipu (Z.ai) founder and Tsinghua professor Tang Jie published an internal memo arguing frontier AI must stay "as open and widely accessible as possible" — "real safety comes from broad participation, sharing, and oversight, not from technological barriers" — and reaffirming GLM-5.2 under an MIT open-source license, committing Zhipu to two years without short-term app monetization.