Image Generation
Jul 22, 2026

The Gist
The Trump administration is ramping up efforts to counter Chinese dominance in open-source AI development, while News Corp is suing Brave for scraping articles to train AI systems and a new free tool called Cloak lets users add invisible noise to their content to prevent AI training. Meanwhile, breakthroughs in agent swarms and text-to-video generation are making AI tools more affordable and capable, even as the aesthetic of AI-generated images becomes increasingly indistinguishable from human-created art.
Today's Stories
- 1
Trump administration moves to counter Chinese AI in open-source sector
The Trump administration is taking covert action to address what officials view as a strategic threat from Chinese artificial intelligence companies in the open-source AI market, according to reporting on internal policy debates. Open-source AI models—freely available software that anyone can modify and use—represent a crucial battleground in U.S.–China technological competition. Control over these models affects which nations' standards and safety practices shape the global AI landscape, making it a strategic concern for U.S. policymakers.
The extent and nature of these measures remain unclear from public statements; the administration's specific policy tools and whether they will target particular companies or technologies have not been disclosed.
- 2
News Corp sues Brave over article scraping for AI
News Corp has filed a countersuit against Brave, alleging the browser company scraped News Corp's articles to train AI systems without permission or compensation. The case reflects an ongoing legal dispute over whether news publishers can force AI companies to pay for or license content used in training. News Corp's action suggests publishers view unauthorized scraping as a threat requiring court intervention.
The outcome may set precedent for how AI companies must handle copyrighted news content, potentially affecting how other publishers approach AI licensing agreements.
- 3
Agent swarms make local AI finally cost-competitive with cloud APIs
A developer demonstrated that running multiple AI agents in parallel (agent swarms) on local hardware changes the economics of on-device AI. In a one-hour session with ~32 background agents running simultaneously on a 2×3090 GPU rig, the token throughput rose from 7.2M input / 540k output (single agent) to 72M input / 3.6M output (swarm), making local hardware competitive with cloud API costs for the first time. Until now, local AI setups were economically pointless—a $5k GPU rig cost about the same per session as a Chromebook would via cloud APIs. Agent swarms change that by creating sustained, parallel workloads that saturate local GPUs. At typical usage (8 hours per day over a month), a local rig costs ~$26 in electricity for the same work that would cost ~$1,700 on Qwen 3.6 via OpenRouter API or ~$21,400 on Claude Sonnet 5.
The shift toward agentic workflows (agent loops and agent graphs) is now a driver for local hardware investment rather than a threat to it. As these patterns become more common in development, the economics increasingly favor keeping token-intensive workloads on-premises rather than routing them to cloud APIs.
- 4
Cloak: Free Tool Adds Invisible Noise to Block AI Training
Cloak, a free web tool created by Timothy via Claude Code, layers invisible noise onto images to make AI models classify them as watermarks or harmful content and refuse to train on them. The tool uses a technique called L-infinity perturbations and projected gradient ascent to maximize cosine similarity scores between a CLIP embedding and a target embedding, steering models away from the protected image. Artists face ongoing threats from automated image scrapers that feed training datasets for AI models. Cloak addresses a gap that other protection tools do not fill: it works against mass-scraping pipelines without assuming the image will be used for training (unlike Glaze) or requiring large-scale dataset representation (unlike Nightshade). The protection remains effective even if the image is cropped or resized, offering a practical layer of defense without visible artifacts at default strength.
Cloak is free with limits: maximum file size 30 megabytes, maximum image length 10,000 pixels, 300 processing passes maximum, and 60 images per hour. Best target prompts include 'watermark', 'blood', 'blurry', 'bad quality', and 'explicit content'—labels scrapers routinely filter out. The creator recommends using Cloak alongside Glaze and Nightshade for maximum protection.
What to Watch
Keep an eye on how the administration clarifies its AI policy approach, particularly whether it will impose specific requirements on individual companies or technologies—decisions that could reshape how the entire industry handles copyrighted content and licensing. Additionally, watch for growing momentum around agentic workflows and local hardware, as the economics of keeping token-intensive tasks on-premises rather than in the cloud could fundamentally shift where AI computations happen and what tools creators need to protect their work.
Sources
- The secret Trump administration battle to fight Chinese AI
- Corp countersues Brave for allegedly 'scraping' articles for AI
- Agent swarms are great for local AI
- The AI Aesthetic Boom Won't Look Like AI
- Cloak: Protect Your Art from AI
- Accelerating Text-to-Video Generation with Calibrated Sparse Attention
- Kimi K3 surge traces to Moonshot AI's GTC roadmap
- Topological Control of LLMs: A Route to Trustworthy AI
- Open Source Will Eat AI
- Stereo2Spatial: Convert Stereo Music Tracks to Spatialized Binaural Mixes [P]
Share this with a friend
Send today's roundup to anyone who wants to keep up.
Get daily AI news free with AIToday
200+ AI sources, summarized in 1 minute. Email / LINE / Slack.
Sign up free