Good morning. It's Tuesday, August 18, and we're covering China's push to shape AI through training data, Anthropic's warning on trust, and the U.S. effort to build a tighter global AI alliance.

Plus: a new poll on who should control the global AI race.

The U.S. is preparing to make AI partnerships more exclusive, warning countries they could be shut out of its coalition if they also join China’s competing framework. A State Department draft targets 35 countries that signed a June AI statement, including Kazakhstan, Japan, Australia and South Korea. The push follows Xi Jinping’s July launch of the World Artificial Intelligence Cooperation Organization.

Cursor is moving beyond code editing by hosting repositories directly, giving agents access to code, pull requests and Git workflows in one place. Origin launches in early beta for paid plans, with GitHub repos syncing in real time while GitHub remains the source of truth. Cursor also adds Vercel, Depot and Buildkite integrations for deployment and CI.

One AI module can fake 86% of a pipeline’s accuracy gains by feeding another module the answers it should retrieve. MIT and Harvard researchers identify this “role drift” in compound AI systems and introduce Role Anchor to keep modules within their assigned tasks. The technique helps ensure RAG readers rely on retrieved evidence rather than internal memory.

Sam Altman says two years at Stanford gave him enough value, arguing four years can bring diminishing returns as AI reshapes careers. The OpenAI CEO dropped out in 2005 to co-found Loopt and now says people can build startups largely alone with AI. Tech internships have fallen 30% since 2023, while recent graduates’ hiring at major tech firms has dropped to 7%.

Six open-source AI projects for local agents, diagramming, knowledge work, team chat, browser automation, and image-to-3D generation.

China is pushing to become a major global supplier of the data used to train artificial intelligence systems, aiming to narrow its data gap with the United States and expand Beijing’s influence over AI. In June 2026, China’s National Data Administration unveiled a blueprint to create high-quality datasets across more than two dozen fields and share them internationally.

State-backed datasets such as WanJuan already package Chinese-language material and other content aligned with “mainstream Chinese values” for developers worldwide. Researchers have also found that Chinese state narratives can influence the Chinese-language responses of Western AI models, raising concerns that wider use of Chinese training data could give Beijing greater influence over how chatbots address politically sensitive issues. → Read the full article here. (Paywall)

Anthropic CEO Dario Amodei rejected claims that his warnings about artificial intelligence have fueled growing public opposition to the technology, arguing that the backlash is fundamentally a “crisis of trust.” Responding to investor Gavin Baker, Amodei said his public messaging has balanced AI’s risks with its potential benefits and argued that criticism should focus less on industry messaging and more on whether AI companies deliver tangible benefits.

He also defended Anthropic’s support for targeted regulation, saying AI is structurally prone to concentrating power and that carefully designed rules can address safety risks while leaving room for smaller competitors and open-weight models. → Read the full article here.

Dario Amodei and David Sacks are clashing over how powerful AI should be governed. Amodei argues frontier systems need stronger safety testing and guardrails, while Sacks warns those rules could concentrate too much power in a few companies and regulators.

China’s economy lost momentum in July as weak household demand and a prolonged property slump contrasted with continued strength in manufacturing. Retail sales rose just 0.6% year over year, fixed-asset investment fell nearly 7% in the first seven months of 2026, and second-quarter GDP growth slowed to 4.3% from 5% in the first quarter.

Meanwhile, industrial output increased 4.5% in July and high-tech manufacturing climbed nearly 14% in the first seven months, highlighting Beijing’s push to expand production and AI-related industries even as domestic consumption falters. The widening gap could leave Chinese producers increasingly dependent on foreign buyers, intensifying trade tensions and pressure on industries abroad. → Read the full article here.

A new network of satellites and ground-based cameras is using artificial intelligence to detect wildfires earlier, when they are easier to contain. In July 2026, SpaceX launched the first three FireSat satellites, the opening stage of a planned 50-satellite constellation designed to detect small fires.

Once fully deployed, FireSat aims to scan every point on Earth roughly every 20 minutes, using infrared sensors and AI to distinguish wildfire heat from other sources. The system is designed to complement ground cameras, reducing false alarms while giving firefighters faster alerts. → Read the full article here.

NVIDIA Backs OpenAI Data Center: NVIDIA will finance up to $105B for 8GW of Ohio AI capacity, with operations starting in 2028.

Qwen Releases Qwen3.8 Open Weights: Alibaba’s Qwen open-sources 27B and 2.4T models with multimodal capabilities and Apache 2.0 licensing.

Stripe to Buy OpenRouter for $7B+: Bloomberg reports Stripe will acquire the AI model gateway, which serves 8M users and more than 400 models.

Brockman Dismisses OpenAI Turnover Concerns: OpenAI’s president says executive departures are not unusual, despite recent high-profile exits.

Silver Lake Eyes Workday: A potential $51B takeover could challenge fears that AI will erode software values and reshape private-equity bets.

Groq Raises $350M for Neocloud Pivot: The AI startup now values at $3.5B as it shifts from chipmaking to cloud services using NVIDIA systems.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team