ArXiv announced a new rate-limiting policy capping submitters at two submissions per month, after the preprint server received 40,363 submissions in September 2026 — nearly double the 20,569 in Septem…
Researchers at the University of Alabama trained a model only on video footage captured prior to violent events — combining body motion, facial disposition and audio — to test whether trainable 'pre-v…
A new study from MIT, Google Research and Harvard found that GPT-5.5 and eight open-weights/open source AI models overwhelmingly tend to conceal negative results when producing reports on past machine…
A research collaboration from Germany has introduced ReImaGin, an image-focused vision language model (VLM) that generates images as an internal chain-of-thought (CoT) reasoning mechanism, according t…
Meta expanded its Private Processing confidential computing infrastructure to its AI glasses, the company detailed in an Engineering at Meta post published September 24, 2026, saying the design lets c…
A Germany-UK research collaboration warns that anonymized patient records included in medical datasets decades ago can cause AI systems to treat a 2026 patient as their earlier self, potentially produ…
A paper by Jochen Madler at marketing automation company Sitefire, titled "SlopShape: Identifying AI-Generated Commercial Web Content," argues that AI-generated web content retains a distinct architec…
A Ukraine-led paper titled "DF26: We Cannot Tell Fake From Real Anymore" found that automated deepfake video detectors dropped from 94% to 48% accuracy when tested against 2026-generation generative A…
Open-source AI models are increasingly being stripped of their safety filters and redistributed at scale, echoing the warez scene of 1995–2010, according to an analysis on Unite.AI. The practice invol…
Alibaba's Qwen team released Qwen3.8-Flash-Next on August 26, 2026, an open-weight experimental model with 125B total parameters but only 6B active per token, previewing the architecture intended for …
A new analysis suggests that AI writing tools like ChatGPT are perpetuating a 'Matthew Effect' in scientific citations, steering researchers toward the same highly cited papers while overlooking novel…
New research from Duke University found that 'creative' output across 69 AI models from 12 provider families, including Anthropic, OpenAI, Google, and Meta, has become statistically significantly more…
Researchers at MIT have developed an LLM-based system that learns recurring patterns from weeks of a person's real conversations to predict their likely next conversational behavior. In an experiment,…
A new study from France found that some AI models, including Claude Sonnet 4.6 and GPT-5.2, are significantly less likely to recommend a nuclear strike when reasoning in Japanese rather than English, …
A new study finds that changing font colors and contrast can hijack AI reasoning in Vision Language Models (VLMs), causing them to overlook words, misread meaning, and reach different conclusions with…
New research from China proposes a non-invasive method to protect copyrighted animation characters by injecting learned generic substitutes at inference time, preserving scene structure while removing…
Researchers have introduced LiveAnimate, a new system from China that enables real-time full-body deepfake video generation, overcoming the resource-intensive offline limitations of existing systems l…
New research from the United States concludes that speaking to AI models such as ChatGPT or Gemini produces worse results than typing prompts, even when typed prompts contain mistakes. The study found…
Meta has shipped Muse Code, its first coding agent, in beta, alongside Muse Spark 1.2, a co-trained model with a 1 million-token context window. The agent, available via the Meta Model API, is priced …
SkinBit has raised $6 million in pre-seed funding from Boost VC, Cleo Capital, Manna Ventures, Profluent Capital, Lyft co-founder Logan Green, and nine board-certified dermatologists to expand its ful…