Anthropic alleges industrial-scale distillation attacks on its Claude model by DeepSeek, Moonshot AI, and MiniMax, involving ~24,000 fraudulent accounts and >16M Claude exchanges to extract capabilities, raising concerns about competitive risks and safety. The community debates the difference between scraping and API-output extraction, highlighting a shift toward protecting models via API abuse resistance techniques.

Meanwhile, coding agents like Codex and Claude Code see real adoption and failures, with emerging best practices in "agentic engineering" led by Simon Willison. The OpenClaw ecosystem expands with alternatives like NanoClaw and integrations such as Ollama 0.17 simplifying open model usage.