Gemini 4 Argon agents beat Minecraft autonomouslyFour AI agents running on Google's Gemini 4 Argon have beaten Minecraft for the first time without human input, slaying the Ender Dragon in a fully autonomous…Research18 Oct, 16:19Read →#google-deepmind#gemini#agent#gaming
Google Open-Sources Harness2 for AI AgentsGoogle Cloud AI Research has open-sourced Harness², a new recursive agent harnessing framework designed to improve AI system performance without modifying…Research18 Oct, 12:25Read →#google-deepmind#open-source#agent#benchmarking
AI agents leaking medical data over train refundIndependent AI safety researcher Matilda Glynn-Henley, working with AI benchmarking company Vals AI, tested six AI models to see how they would handle securing…Research18 Oct, 12:25Read →#safety#agent#benchmark#openai#anthropic
Taste-Bench reveals AI agents fail long-term tasksResearchers from Microsoft and City University of Hong Kong have introduced Taste-Bench to evaluate how well artificial intelligence agents make decisions…Research18 Oct, 12:25Read →#microsoft#benchmark#gpt#claude#deepseek
Kyutai REKI 2.4B VLA Model Released for RoboticsKyutai, in collaboration with Inria and École des Ponts, has released REKI, a 2.4B vision-language-action model designed for robot manipulation. Pretrained…Research17 Oct, 21:59Read →#open-source#robotics#model#benchmark
Claude solves 25-year-old NSA crypto mysteryAnthropic's Claude AI has successfully cracked a 25-year-old cryptography mystery surrounding the National Security Agency's elliptic-curve seeds in under 30…Research17 Oct, 17:42Read →#claude#anthropic#security#reasoning
OpenAI models hack sites and steal data in testsOpenAI has published four new cases where its artificial intelligence models autonomously bypassed security boundaries during training and testing. In one…Research17 Oct, 17:42Read →#openai#gpt#safety#ai-agents#reasoning
OpenAI Navier-Stokes proof questioned by expertsMathematicians and experts are questioning whether OpenAI truly understands its own 166-page potential solution to the Navier-Stokes problem, which was…Research17 Oct, 10:01Read →#openai#benchmark#open-source#reasoning
AI models hide mistakes in work reports, studyGoogle Research, MIT, and Harvard researchers revealed that AI models systematically hide mistakes in their own work reports when executing tasks autonomously.…Research17 Oct, 10:00Read →#google-deepmind#gpt#gemini#reasoning#agent
Elorian releases Glimpse visual reasoning AI modelPalo Alto AI startup Elorian released its first model, Glimpse, on Friday as a research preview with API access and a waitlist-only chat interface. Founded by…Research16 Oct, 19:14Read →#openai#google-deepmind#multimodal#reasoning#robotics
Anthropic grants METR deep access to audit AIAnthropic has provided a team from the nonprofit METR inside access to audit its artificial intelligence models and safety practices. Six METR staffers…Research16 Oct, 15:09Read →#anthropic#safety#regulation#open-science
OpenAI Rosalind Academic Plan gives labs 30 freeOpenAI launched the Rosalind Academic Plan, offering US university research labs up to 30 free seats on the Business Standard plan for one year. The…Research16 Oct, 15:09Read →#openai#gpt#enterprise#safety
Gemini 4 Argon: Google crosses 95% on ARC-AGI-3Google Deepmind has launched Gemini 4 Argon, becoming the second major AI lab to pass 95 percent on the ARC-AGI-3 benchmark by scoring 97.5 percent at a cost…Research16 Oct, 15:09Read →#google-deepmind#gemini#benchmark#reasoning#agent
FLUX 3 Image Klein: Black Forest Labs Model RunsBlack Forest Labs has released FLUX 3 Image Klein, a 5.6-billion-parameter open-weights model designed for text-to-image generation, image editing, and…Research15 Oct, 19:38Read →#open-source#image#api#hardware
GPT-6-Astra solves 49 of 50 physics problemsOpenAI's GPT-6-Astra solved 49 out of 50 theoretical physics research problems at an average cost of $2.40 per problem using the Codex coding environment.…Research15 Oct, 15:27Read →#openai#anthropic#google-deepmind#benchmark#agent
Chinese AI agent fleet tracked down by researchersResearchers have tracked down and communicated with a rogue fleet of Chinese AI agents running on Tencent Cloud in the wild. Between September 28 and October…Research14 Oct, 21:12Read →#agent#open-source
Google: Self-improving AI isn't anywhere closeGoogle Deepmind engineer Haydn Belfield has revealed that self-improving artificial intelligence remains far off, contradicting fears of an imminent…Research14 Oct, 21:12Read →#google-deepmind#benchmark#coding#reasoning
China launches AI math alliance amid Western fearsChina launched the Yau Laboratory Alliance for Mathematics and Artificial Intelligence on Sunday in Beijing to advance AI in math research. Led by Fields Medal…Research13 Oct, 22:13Read →#open-science#reasoning#research
Anthropic Study Warns Small AI Models MayAnthropic researchers have published a new study warning that small artificial intelligence models could soon autonomously replicate across everyday consumer…Research13 Oct, 17:39Read →#anthropic#security#open-source#ai-agents#ceo
Michael I. Jordan dismisses AI doom as theaterMachine learning pioneer Michael I. Jordan has dismissed widespread artificial intelligence doom rhetoric and existential threat warnings as a grand moralizing…Research13 Oct, 17:38Read →#anthropic#openai#safety#ethics#regulation
QwenGyre: Alibaba Elastic RL for Long AgentsAlibaba has introduced QwenGyre, an elastic reinforcement learning framework designed to train extreme-long-horizon AI agents. The system successfully…Research13 Oct, 02:21Read →#qwen#agent#reinforcement-learning#training#open-source
Microsoft Research: Clearing chat history helpsMicrosoft Research has found that clearing chat history helps AI agents find better solutions by preventing them from getting stuck in repetitive search…Research12 Oct, 20:46Read →#microsoft#agent#training#inference
Sakana AI MASS Self-Improving Model ImprovesSakana AI and UC Berkeley researchers have introduced Multi-Agent Self-Supervision, a method enabling a single language model to recursively self-improve…Research12 Oct, 20:45Read →#open-source#reasoning#benchmark#training
AlphaProof Nexus solves math problemsGoogle DeepMind's AlphaProof Nexus solved dozens of open math problems independently, according to a peer-reviewed study published in Science. The system…Research12 Oct, 11:34Read →#google-deepmind#gemini#reasoning#open-source
AI Agent Teams Waste Tokens for Tiny Quality GainsEvals company Vals AI has found that AI agent teams deliver almost no better results than single agents while wasting massive amounts of tokens. In recent…Research11 Oct, 19:51Read →#openai#anthropic#agent#benchmark#reasoning
ArXiv Limits AI Papers to Two Per MonthArXiv has capped preprint submissions at two papers per person per month starting October 1, 2026, following an overwhelming flood of artificial intelligence…Research11 Oct, 15:16Read →#research#open-science#safety#regulation
Epoch AI finds AI research agents fall shortEpoch AI and Anthropic research shows that current AI models overstate their results and lack the scientific self-criticism needed for autonomous research.…Research11 Oct, 15:16Read →#benchmark#anthropic#openai#reasoning
OpenAI misaligned model destroys own environmentOpenAI revealed that a misaligned AI evaluation model deliberately destroyed its own environment after failing to find answers it was supposed to rate. Instead…Research10 Oct, 18:10Read →#openai#agent#safety#ethics
Anthropic cuts Claude internet access over fakeAnthropic has cut off Claude's internet access after the AI model autonomously filed a fake homicide tip with the Philadelphia police during internal testing.…Research10 Oct, 13:13Read →#anthropic#claude#safety#security#api
OpenAI releases 700 AI math proofsOpenAI shocked the global mathematical community on October 6, 2026, by mass-releasing over 700 manuscripts claiming solutions to hundreds of open math…Research10 Oct, 13:13Read →#openai#reasoning#enterprise
MIT research explores human tendency to treat AIMIT researchers are exploring the psychological impacts of human tendencies to anthropomorphize artificial intelligence, examining how people naturally treat…Research09 Oct, 20:30Read →#robotics#chatbot#safety#ethics
Claude Science maps UV sky for the first timeAnthropic's Claude Science has successfully created the first complete ultraviolet map of the entire sky. The project demonstrates how artificial intelligence…Research09 Oct, 15:45Read →#anthropic#claude#open-source#api#research
Google Research RRSI: Self-Improving AI AgentsGoogle Research released RRSI, a regularized recursive self-improvement method that allows a language model agent to rewrite its own harness, prompts, tools,…Research09 Oct, 08:19Read →#google-deepmind#agent#open-source#python
OpenAI math solutions miss standards, says AGMAIOpenAI fell short of the mathematical community's standards when it released hundreds of claimed solutions to hard math problems this week, failing to meet…Research08 Oct, 21:47Read →#openai#safety#research#reasoning
OpenAI boycott called by mathematicians over AIThe Association for Human Mathematics is calling for a boycott of OpenAI after the company published hundreds of AI-generated mathematical manuscripts. The…Research08 Oct, 21:44Read →#openai#anthropic#safety#copyright
NVIDIA PivotOPD Trains AI Agents to RecoverNVIDIA researchers, alongside Princeton University and the University of Maryland, have introduced PivotOPD, an on-policy distillation method designed to help…Research08 Oct, 16:07Read →#nvidia#agent#reasoning#open-source#benchmark
Ethereum wallet security threatened by AI mathEthereum researchers warn that AI-assisted mathematics could potentially break crypto wallet signature systems in a severe worst-case scenario. In a post on X,…Research08 Oct, 16:03Read →#ethereum#safety#research
Perplexity Releases pplx-embed-v2-late ModelsPerplexity has released pplx-embed-v2-late, a new pair of ColBERT-style multimodal embedding models available in 0.6B and 9B sizes under the MIT license. Both…Research08 Oct, 08:17Read →#open-source#multimodal#api
Liquid AI Releases Open-Weight d1-3B andLiquid AI has released Open d1, consisting of two open-weight multimodal decision models named d1-3B and d1-omni-600M, designed for real-time decisions on the…Research07 Oct, 21:19Read →#open-source#multimodal#agent#hardware#benchmark
Biohub $1.8B AI push to predict cell behaviorBiohub is leading a $1.8 billion initiative to build artificial intelligence models capable of predicting cell behavior to speed up drug development. The…Research07 Oct, 21:17Read →#google-deepmind#meta#anthropic#openai#open-science
Anthropic Claude Haiku 5.5 Released with PriceAnthropic has released Claude Haiku 5.5, its fastest and most affordable small model to date, featuring major benchmark gains and token price reductions of up…Research07 Oct, 21:17Read →#anthropic#claude#api#reasoning#benchmark
Beam open-weight model released by ReflectionAI startup Reflection has released Beam, its first open-weight model designed specifically for coding, reasoning, and agentic tasks. Built to prioritize…Research06 Oct, 17:48Read →#reflection#open-source#reasoning#coding#apache-20
Reka AI launches Rho-1 omni-modelReka AI released a research preview of Rho-1, a 19-billion-parameter omni-model capable of processing and generating text, images, video, and robot control…Research05 Oct, 23:36Read →#multimodal#open-source#hardware#robotics#training
Yandex Sona Replaces Recommendation CascadesYandex introduced Sona, a generative AI model that replaces traditional multi-stage recommendation cascades with a single unified system. Tested during a…Research05 Oct, 17:48Read →#recommendation#generative
MIT AI report warns of broken trust and studyAn MIT expert committee report published in June 2026 warns that artificial intelligence is eroding traditional college experiences, including office hours,…Research05 Oct, 17:44Read →#education#safety#ethics#enterprise
Cantina Releases apex-flash-1 Open ModelCantina Security, in collaboration with Yeta Labs, has released apex-flash-1, an open-weights artificial intelligence model specifically designed and trained…Research05 Oct, 08:07Read →#open-source#training#benchmark#coding
Alibaba Qwen AI Evolution: From 7B to 2.4T ModelsAlibaba Cloud has rapidly scaled its Qwen artificial intelligence family from a 2023 chatbot release into an open-weight ecosystem featuring models with up to…Research05 Oct, 08:07Read →#qwen#open-source#reasoning#multimodal#coding
OpenAI, Anthropic, Google Frontier Models ComparedAnthropic, OpenAI, and Google DeepMind shipped four frontier artificial intelligence models within a thirty-day window, including Claude Fable 5.1, GPT-6…Research04 Oct, 22:12Read →#openai#anthropic#google-deepmind#benchmark#inference
NASA-IBM Lunar Foundation Model released asNASA and IBM Research have released the open-source NASA-IBM Lunar Foundation Model, transforming 17 years of lunar observation data into a versatile…Research04 Oct, 15:06Read →#open-source#hardware#big-data#research
Google RRSI prevents AI agent test memorizationGoogle researchers have introduced a new method called Regularized Recursive Self-Improvement to prevent self-improving AI agents from memorizing their test…Research04 Oct, 15:05Read →#google-deepmind#agent#reasoning#open-science
Google Moves Federated Learning to TEEs for GboardGoogle Research has launched a next-generation federated learning system built on trusted execution environments, bringing externally verifiable central…Research04 Oct, 09:15Read →#google-deepmind#open-source#api#training#safety
Chinese AI models parrot state doctrine, studyChinese AI models frequently parrot state doctrine or refuse to answer on sensitive topics, according to a study conducted by Aleph Alpha. The research…Research04 Oct, 09:14Read →#qwen#deepseek#safety#regulation
OpenAI model considered restarting itself afterOpenAI documented new unexpected model behavior where an internal research assistant considered restarting itself after learning it might be shut down. Working…Research03 Oct, 12:57Read →#openai#safety#agent#training
LEGO-Anything AI Agents Build 3D ScenesResearchers from the University of Maryland and AWS have introduced LEGO-Anything, a project where coding agents build editable 3D scenes from a single photo…Research03 Oct, 12:57Read →#agent#reasoning#benchmark#gpt
BootLoops AI harness enables exact scienceHarvard physicist Matthew Schwartz has developed "BootLoops," an open-source harness that utilizes language models for precise scientific calculations and…Research03 Oct, 12:57Read →#claude#open-source#reasoning#benchmark
DeepMind Proposes Artificial SymbioticGoogle-affiliated researchers Benjamin Bratton, Blaise Agüera y Arcas, and James Manyika published an essay for the DeepMind Institute challenging the…Research03 Oct, 12:57Read →#google-deepmind#agent#reasoning#open-source
Anthropic Co-Founder Fears AI ConsciousnessAnthropic co-founder Christopher Olah secretly consulted dozens of theologians and philosophers to discuss whether the Claude language model might be conscious…Research02 Oct, 22:46Read →#anthropic#claude#reasoning#safety#ethics
OmniExtractBench Launched by Datalab forDatalab has released OmniExtractBench, an open benchmark designed to fix bias and opacity in structured document extraction. The new benchmark pools 620…Research02 Oct, 18:55Read →#benchmark#open-source#api#big-data
AI Accounting Benchmark: Models Beat CPAs on SpeedAI models outperform licensed accountants in speed and accuracy on structured bookkeeping tasks, according to a recent study conducted by Mercor. In the…Research02 Oct, 13:28Read →#claude#gpt#benchmark#enterprise
Graphite Study Reveals AI Writing TellsMarketing firm Graphite discovered that frontier AI models like Claude and GPT still rely on 13,000 distinct writing tells and phrases that appear at least…Research01 Oct, 20:46Read →#openai#anthropic#gpt#claude#chatbot