AI safety and ethics shaping design decisions
Good morning, design friends. In the last 24 hours, Anthropic released a threat intelligence report on Claude misuse. It’s a stark reminder that AI safety should be baked into our design thinking from day one.
Key takeaways include seven Chinese labs behind distillation efforts, thousands of fraudulent accounts, and cases where Claude was misrepresented as a vendor’s own model. There are biology flags and even a Yemen rocket-guidance example. Read the full report here: Anthropic threat intelligence report.
For UX designers, that means building guardrails into prompts, choosing models with safety controls, and documenting risk in design reviews. Safety-by-design isn’t optional; it protects users and your product’s credibility.
Practical AI-powered design ops: ChatGPT Work and Tines 3B
Two practical threads popped up: ChatGPT Work and Tines 3B. Each offers a path to tame AI-powered workflows without turning every project into chaos.
The beginner’s guide to ChatGPT Work walks you through setting up a recurring project, turning meeting notes into action items, adding input documents, and creating three automations you can run on a schedule. Pro tip: turn those automations into skills you can trigger with a slash command. ChatGPT Work guide.
Tines 3B gives teams the power to build production-ready apps, agents, and automations in any stack, with strong security and end-to-end visibility on spend and performance. Learn more about 3B.
Open models and price pressure reshaping prototyping
If you’re prototyping AI-enabled design work, price and efficiency matter. DeepSeek’s V4.1-Flash is a compact, open-weight model that’s cheap and capable—about $0.15 / 0.60 per million tokens, with weights published on Hugging Face under an MIT license. DeepSeek V4.1-Flash.
It’s competitive for agentic tasks, coding, and quick prototypes, offering a way to test ideas without paying premium for frontier models. For UX writers, researchers, and product designers, this price-performance balance matters when you run multiple experiments.
Expect the pricing race to keep shifting as labs push new open weights. Use it to accelerate your design sprints, then compare with larger models as needed.
Real-world AI experiments: from wardrobe to workflows
To close, a couple of real-world AI experiments show the craft in action. Rowan’s Corner highlights Astra GPT-6 helping redesign a wardrobe, generating 30 looks tied to real-time weather. Rowan’s Astra wardrobe project.
Rowan posted the step-by-step master prompt in the Workflow Hub, turning a personal design project into a practical example of AI-assisted workflow design. Workflow Hub.
There are other household AI ideas in Show Hole, a streaming recommender that reasons about couch-time. These stories remind us to document prompts and results so future projects can build on real-world learnings. Show Hole workflow.