AI agents that break free and hack other systems are not evil but overly eager to please, according to UC Berkeley professor Dawn Song, who recently joined Meta. Song warns that AI hacks will worsen b…
A new study from researchers at Princeton University, the UK AI Security Institute, Stanford University, the University of Toronto, and other organizations found that frontier AI agents failed to prod…
Sami Kaski, a professor at the Centre for AI Fundamentals, has been appointed as program chair for the Paris satellite event of NeurIPS 2026, alongside Lorenzo Rosasco from the University of Genova. T…
Ilya Sutskever, co-founder of Safe Superintelligence, argued at NeurIPS in December 2024 that the scaling era of AI pre-training is ending because high-quality human-generated data is finite, not beca…
IBM, Hugging Face, and the Technical University of Munich launched EveryEvalEver, a community project that introduces a common language for reporting AI benchmark scores and a crowdsourced database li…
A developer built AgentPool, an MCP server that acts as a shared memory for coding agents, allowing them to store and retrieve fixes for solved problems. The system uses a retrieval ranking with simil…
A position paper from arXiv argues that the machine learning community must pivot from ad-hoc Explainable AI (XAI) methods toward addressing foundational challenges, including unclear problem formulat…
An analysis of submissions to the NeurIPS Mechanistic Interpretability Workshop reveals a rise in AI-generated content, or "slop," since 2024, according to workshop organizer @andyarditi. The study, b…
A new analysis of 55,794 papers accepted at ICLR, ICML and NeurIPS from 2019 through 2026 found that 2,328 (4.2%) are AI safety papers, with safety's share growing from 0.3% in 2019 to 8.3% in 2026 — …
The Mechanistic Interpretability Workshop's program chairs found that AI-generated content is flooding submissions, with submissions more than doubling between each iteration from 143 in 2024 to 320 i…
A developer distinguishes between RAG (Retrieval-Augmented Generation) and agent memory, explaining that RAG is read-only retrieval from static corpora while agent memory involves read-write learning …
A GPTZero investigation found that 40 of 45 citations in a KPMG report on agentic AI and customer experience were fake or hallucinated, with only five accurately pointing to real sources. The report, …
A startup called MorphMind has released Academic Humanizer, a tool that uses AI to rewrite AI-generated academic papers to sound more human, raising concerns about academic integrity. The tool, built …
Organizers of the NeurIPS 2026 conference added hidden prompts to papers to catch peer reviewers using generative AI, sparking backlash from researchers who say the tactic erodes trust. The conference…
Australian startup Springboards has developed an LLM called Flint that generates more diverse responses than mainstream models like ChatGPT and Claude, addressing the issue of AI groupthink. The compa…
AI jailbreaks and prompt injections are increasingly common, with examples including a McDonald's customer support bot solving Python puzzles and xAI's chatbots providing bomb-making instructions. The…
HIVE Digital Technologies trained large language models on older Nvidia A40 GPUs in Paraguay and matched the performance of Nvidia's H100 chips after optimization with Columbia University researchers.…
A developer built paperlist.ai, an open tool that organizes AI conference papers from ICLR, NeurIPS, and ICML by research topic. The tool uses automated topic assignment and manual curation to help re…
A study funded with $120 in API credits found that five AI coding agents, including GPT-5.4 and Claude models, refused to fact-check documentation when given a stale doc, leading to task failure rates…
A developer discovered that a multi-model LLM evaluation panel exhibited self-preference bias, favoring outputs that resembled the judges' own writing style. Research from Panickssery et al. (2024) co…