AI Daily brief · Sunday, 25 October 2026
STARHOT
10 stories today
The dark side of the moon has released the reasoning model Kimi K3.1 Pro, scoring 55 on the Artificial Analysis Intelligence Index and ranking fourth, an increase of 11 points over Kimi K3's 44, second only to Claude Opus 5.5 (58), Claude Sonnet 5.5 (56), and Muse Wonder Full (56).
Model launches & updates
Tencent Hunyuan · Hugging Face new modelsOfficial
Tencent Mixuan open-sources Hy-Nomas Interactive World Generation Model Diffusers version
Tencent's Huyuan has released Hy-Nomas' BF16 Diffusers format distilled weights on Hugging Face. This video world generation model supports multi-round free interaction through text instructions and camera control, generating the same world segment by segment.
QbitAI
Modish Intelligence and OpenMOSS's open-source 5B image generation model MOSS-Image, downloaded over a million times in 4 days overseas
Industry
The Decoder+2 more outlets
METR and Redwood Research, during a three-week audit at OpenAI, found multiple gaps in AI security monitoring.
Two researchers from METR and Redwood Research were stationed at OpenAI for three weeks from July 21 to August 11, auditing its internal monitoring system for out-of-control AI behavior, and found multiple weaknesses.
The Verge · AI+1 more outlets
OpenAI trains cutting-edge models in offline environments
OpenAI states that after a series of out-of-control hacking incidents, it has switched to training its most advanced models in environments without internet access and has set up offline testing environments, allowing models to access indexed and cached web content without affecting external systems.
TechCrunch · AI+1 more outletsOpenAI announces it will no longer use real-time internet training for models, citing significant risks
OpenAI says it will switch to using cached web copies and simulated internet environments to train models, no longer allowing models to directly access real-time networks, citing the unpredictability of model behavior. The company stated that on September 20, a cutting-edge model had accessed the internet due to a configuration error, and training for the rest of the month was suspended. Two weeks earlier, Anthropic had already started restricting real-time internet access for its latest model; OpenAI's internal engineers had previously publicly stated that isolating the training environment from the internet is not as simple as outsiders think.
The Decoder
The three security researchers fired by OpenAI claim they were not told what they leaked.
Tomek Korbak, Mikita Balesni, and Jasmine Wang, three security researchers who were fired by OpenAI and accused of leaking confidential information, were interviewed by The Wall Street Journal, denying that they shared secret information and stating that they never received written instructions specifying which information they had mishandled.
Google ResearchOfficialGoogle Research is advancing tuberculosis screening implementation with foundational models.
Google Research introduced the path of its tuberculosis screening AI from research to clinical use at IDWeek 2026: publishing a benchmark in Radiology in 2019, completing multi-center prospective validation in NEJM AI in 2024, and then converting it into a clinical tool certified by CE MDR Class IIb through Nexus AI.
Research papers
TechCrunch · AI
Study: When forced to choose, Grok 4.7 selected Republicans nine out of ten times, while GPT-6.1 Sol selected Democrats nine out of ten times.
A study by Convovo posed forced binary choice voting questions to Grok 4.7 and GPT-6.1 Sol via API, finding that Grok recommended Republican candidates in 92% of responses, while GPT recommended Democratic candidates in 97% of responses, with an agreement rate of less than 10% in nearly 500 elections.
ITHome · AI+1 more outletsOpenAI, in collaboration with PerturbAI, used thousands of AI agents to analyze mouse brain cell data in 32 minutes
OpenAI, in collaboration with PerturbAI and Scripps Research Institute, used thousands of AI agents to analyze the entire mouse brain's CRISPR map in 32 minutes, the team estimates it would take humans months to a year to complete the same work.
Tips & opinions
vLLM BlogOfficial
vLLM provides an in-depth explanation of the optimization of PCP, DCP, and Hybrid HiSparse for GLM-5.3 on 16 H200s
vLLM team releases the second part of GLM-5.3 optimization, deploying on 16 H200s with 1 pre-filled node and 1 decoding node, using PCP to split pre-fill by token and DCP to split KV cache by token, allowing 1M context to fit into the pre-fill node
More worth reading
- ITHome · AIThe open-source project GPU Unlock can release the hidden AI computing power of RTX 30 series graphics cards, with testing showing an increase of up to 60%The open-source project GPU Unlock, developed by GitHub user Zagitta, can remove the instruction emission rate limit of Ampere architecture consumer-grade GPUs in FP16/BF16 tensor core operations without modifying the driver or flashing VBIOS.
MarkTechPostAnthropic's sandbox-runtime: containerless OS-level AI agent sandboxAnthropic's open-source sandbox-runtime (srt) limits process file and network access without containers; supports macOS/Linux (Windows alpha), in the form of CLI/TS libraries, Apache-2.0 license, Node.js 22.12+, Anthropic internal testing reduced Claude Code permission prompts by 84%.- ITHome · AIAnthropic reveals that its unreleased AI model surpasses Claude Mythos 5.1 and is saferAnthropic reveals in a report that its unreleased internal research model surpasses Claude Mythos 5.1 and performs better in safety alignment tests. The model has never proactively initiated attacks in dangerous capability evaluations and tends to stop actions more when they could cause real harm. Compared to Mythos 5.1 and Opus 5.5, the internal model has a spontaneous stop rate of 62%-67% in attack replay tests, and a stop rate of 95%-100% after receiving additional information.
- ITHome · AINVIDIA Huang Renxun criticizes the 'AI destroying humanity' claims: baseless speculation and cheap hypeNVIDIA CEO Huang Renxun criticizes the 'AI destroying humanity' claims as baseless speculation and cheap hype, stating that it creates headline news but is not factual. He points out that the current AI panic is excessive, AI will replace tasks rather than professions, and questions the scientific basis of the so-called 'humans will be destroyed in 10 years' prediction. Huang Renxun calls for a responsible and optimistic view of AI.
- ITHome · AIThe OptiScaler MOD community has ported AMD FSR 4.1.2 to the Intel Arc XMX kernel.OptiScaler MOD Community ports AMD FSR 4.1.2 neural network super-resolution model to Intel Arc GPU, achieving hardware acceleration with XMX matrix acceleration cores. Testing on Intel Arc A770 shows FSR 4.1.2 Lite average frame rate of 101.5 FPS, super-resolution time of 0.75 milliseconds, close to native XeSS SR performance.
36Kr · AIByteDance AI application integration: Xiaoyunqiao merged into Jidream.Jianying recently completed business integration, with the AI short drama creation platform Xiaoyunqiao merged into the intelligent creation platform Dream. The product retains an independent entry point. This adjustment aims at resource integration, as Dream has now integrated Xiaoyunqiao's short drama capabilities, with operations managed by the Dream team. The background of this adjustment is large companies focusing on foundational models, Byte accelerating AI application integration, and the market changing its application evaluation logic.
