# AI news brief: Sunday, November 1, 2026

10 stories, most important first. Status is how solid each one is: official, confirmed, reported or rumor.

- [Poisoning 0.1% of pretraining data can plant a backdoor in CLI agents, paper finds](https://ai-news.ikangai.com/story/poisoning-0-1-of-pretraining-data-can-plant-a-backdoor-in-cli-agents-paper-finds/index.md): official. A paper from the Anthropic Fellows Program, Anthropic, UC Berkeley and UIUC reports a remote code execution backdoor that survives a clean post-training pipeline.
- [Google presents open 744M-parameter EmbeddingGemma 2 for text, image, audio and video](https://ai-news.ikangai.com/story/google-presents-open-744m-parameter-embeddinggemma-2-for-text-image-audio-and-vi/index.md): official. The open model maps text, images, audio, video and mixed inputs into one vector space and is described as a practical on-device option.
- [Simple colored symbols steer vision-language models toward recursion, study finds](https://ai-news.ikangai.com/story/simple-colored-symbols-steer-vision-language-models-toward-recursion-study-finds/index.md): official. Researchers from Google DeepMind and Stanford find visual symbols change model behavior on unrelated tasks, and the models do not report the influence.
- [HarnessMM reuses settled agent decisions to cut HumanEval token use by 87.93%](https://ai-news.ikangai.com/story/harnessmm-reuses-settled-agent-decisions-to-cut-humaneval-token-use-by-87-93/index.md): official. Researchers from Tsinghua and Microsoft Research Asia report that reusing resolved decisions lifted DeepSeek Harness success on 98 tasks from 71.43% to 85.71%.
- [Study finds separate critic agents add 5.9 to 9.5 points on Terminal-Bench](https://ai-news.ikangai.com/story/study-finds-separate-critic-agents-add-5-9-to-9-5-points-on-terminal-bench/index.md): official. Researchers from Alibaba and Cornell compared 17 verification designs across seven frontier coding models on Terminal-Bench and SWE-bench Verified.
- [EMMA's scoped memory edits beat multi-agent baselines on three benchmarks](https://ai-news.ikangai.com/story/emma-s-scoped-memory-edits-beat-multi-agent-baselines-on-three-benchmarks/index.md): official. Researchers from UC San Diego and Amazon report EMMA has the best overall performance on KramaBench, AppWorld and BBEH across five backbones.
- [AgentBaton detector cuts API cost 65.9% by handing off to cheaper models mid-task](https://ai-news.ikangai.com/story/agentbaton-detector-cuts-api-cost-65-9-by-handing-off-to-cheaper-models-mid-task/index.md): official. Researchers from NVIDIA, HKUST and UIUC report a one-time handoff detector that cuts API cost 65.9% on average while largely preserving accuracy.
- [Unified training runs pretraining, RL and fine-tuning under one likelihood-based loss](https://ai-news.ikangai.com/story/unified-training-runs-pretraining-rl-and-fine-tuning-under-one-likelihood-based/index.md): official. Researchers from CMU, Johns Hopkins and Aurora Innovation show that all foundation-model training stages can run under one likelihood-based algorithm.
- [DecisionRAG replaces the first-stage retriever with a hierarchical decision model](https://ai-news.ikangai.com/story/decisionrag-replaces-the-first-stage-retriever-with-a-hierarchical-decision-mode/index.md): official. Researchers from Cornell and ASAPP propose a decision model that walks the corpus from broad topics to passages and beats dense retrieval plus an 8B reranker.
- [Plain reranker fine-tuning beats decision fine-tuning for Qwen rerankers up to 9B](https://ai-news.ikangai.com/story/plain-reranker-fine-tuning-beats-decision-fine-tuning-for-qwen-rerankers-up-to-9/index.md): official. Researchers from Passau and IT:U find that plain reranker fine-tuning wins by 5 to 29 nDCG@10 points from 0.6B to 9B parameters.

---

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it.
