Anthropic blocks users for anything touching training larger models

More in Artificial Intelligence

• Tests: chatbots rely on three obscure AI-built voter sites for US midterm answers; Copilot cited the error-prone Ballot Brief in 75% of tests and ChatGPT in 22% (Washington Post)

  Major conversational engines are directing election queries to questionable platforms, raising concerns about misinformation during political campaigns. Created by Ben Torres Ezrick out of frustration with existing candidate guides, Ballot Brief has become a primary reference despite recurrent inaccuracies. This reliance highlights vulnerabilities in how automated assistants aggregate and deliver critical civic information to the voting public.

  1 min read
• DeepSeek Harness team published the underlying architecture for their Super-agent, an always-on autonomous agent, as an open-source prototype on GitHub.

  Developers and the broader technology community now have public access to the structural framework behind this emerging system. By sharing the prototype on GitHub, the team enables external researchers to inspect and build upon an autonomous agent designed for continuous operation. This open approach provides immediate visibility into the technical design intended to sustain uninterrupted agent tasks.

  1 min read
• Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard

  Recreating the multi-agent GVS5H architecture inside a new workflow structure demonstrates how collaborative, specialized smaller models can outperform massive standalone systems on difficult coding tasks. The architecture isolates five distinct agents using disk-based messaging and automated test scripts to solve complex benchmark challenges. However, the evaluation noted risks of training data contamination and loop-related execution errors.

  7 min read