ServiceNow introduced MosaicLeaks, a benchmark assessing whether research agents can keep provided confidential information secret. It probes how agents may piece together fragments to infer and leak secrets, exposing weaknesses in their information-handling.
Read the AI / TECH mainstream — extracted from the daily noise.
📊 Recent 7 days
1122
▲ +54%
🔥 1
This Week
In the second week of June 2026, both new models and developer tools surged over +200% week-over-week. Claude Fable 5 hit general availability at the Mythos tier, DiffusionGemma sped generation 4x, and Cohere and NVIDIA pushed agentic coding. A US-ordered pause and swift restoration of Fable 5 also drew focus as frontier models advanced toward deployment and control together.
Top 5 Topics This Week
1
Safety & Evaluation
#Reinforcement Learning
RL agent dev tools tighten around NVIDIA stack
Edge-side reinforcement-learning agents are getting lighter and more memory-efficient, anchored by the NVIDIA JetPack 7.2 release. Agentic-Ready AI is shifting from research to implementation, with academic work running in step. The 49 academic papers at the core surface advances in reward modeling and continual learning optimization, packaged for developers. This week's TOP1 hotspot (57 articles) blends edge inference with long-horizon RL, with Code-language model integration also in scope.
SAMPLEAnthropic opens Seoul office and announces new partnerships across the Korean AI ecosystem
▲ 258 +58%
Sources
Official 8
Academic 239
Specialty 4
Community 7
2
Safety & Evaluation
#Neural Network
Nemotron line reaccelerates neural-net inference stacks
NVIDIA Nemotron 3 Ultra arrives as an inference backbone for long-running agents, pulling neural-network work back toward 'faster + more efficient' as the central theme. Tight coupling with Codex-style agent workflows is now visible.
SAMPLEAnthropic opens Seoul office and announces new partnerships across the Korean AI ecosystem
▲ 233 +54%
Official 8
Academic 208
Specialty 2
Community 15
3
Safety & Evaluation
#Retrieval-Augmented Generation (RAG)
Agentic-RAG infra layer strengthens via DOCA
NVIDIA DOCA In-Silicon Security lands as RAG infrastructure for the agentic-AI era. Amid a steady cadence of new model releases, hardware-level RAG reinforcement is becoming a central technical axis.
SAMPLEResults from the first Anthropic Public Record
▲ 139 +30%
Official 2
Academic 135
Community 2
4
Safety & Evaluation
#AI Agents
Autonomous AI agents move into long-horizon reasoning
Official and academic signals stand at parity, with Nemotron 3 Ultra-backed long-running agents as the standout. AI agents are moving deeper into operational use, with inference cost vs. reliability becoming the dominant tradeoff.
SAMPLEFrom the Hugging Face Hub to robot hardware with Strands Agents and LeRobot
▲ 114 +54%
Official 6
Academic 97
Specialty 9
Community 2
5
Developer Tools
#Deep Learning
Deep Learning
SAMPLEPLaMo-3.0-Prime-β を LLM 開発の現場で使う
▲ 91 +32%
Official 5
Academic 83
Community 3
A. Technical Topics
B. Infrastructure & Operations
C. Industry & Society
Safety & Evaluation
Fable 5 safety fears draw US scrutiny
322
↑ +56%
Developer Tools
SpaceX buys Cursor; Origin Git launches
315
↑ +35%
New Model Releases
GLM-5.2 leads open-weight models
269
↑ +39%
Inference & Efficiency
DiffusionGemma: 4x faster generation
121
↑ +19%
Infrastructure & Hardware
NVIDIA Blackwell tops MLPerf 6.0
111
↑ +35%
Multimodal
Runway in Japan + Video2GUI expand uses
102
↑ +44%
Training & Fine-tuning
Timing trick cuts training energy 14%
100
↑ +22%
Industry Adoption
Hitachi adopts OpenAI Codex for legacy
84
↑ +47%
Funding & M&A
SpaceX acquires Cursor for $60B
74
— Baseline forming
Agents & Tool Use
NVIDIA tops first agentic coding bench
33
— Baseline forming
Policy & Regulation
31
— Baseline forming
AI / TECH activity(volume = articles × weighted mentions)
- ▲ 2026-06-16 — machine_intelligence (z=3.30)
- ▲ 2026-06-16 — openai (z=2.77)
- ▲ 2026-06-17 — openai (z=2.68)
- ▲ 2026-06-18 — openai (z=2.60)
- ▲ 2026-06-18 — gpt (z=2.40)
📰 Featured articles
-
1MosaicLeaks: Can your research agent keep a secret?ServiceNow unveils MosaicLeaks, testing if research agents leak secrets
-
2New usage analytics and updated spend controls for enterprisesOpenAI adds usage analytics and spend controls to ChatGPT EnterpriseOpenAI introduced new usage analytics and updated spend controls for ChatGPT Enterprise, helping organizations track and manage AI costs while scaling with confidence. Admins gain visibility into per-team consumption and can set limits to optimize spend.
-
3Improving health intelligence in ChatGPTOpenAI improves ChatGPT health responses with GPT-5.5 InstantOpenAI says GPT-5.5 Instant strengthens ChatGPT's health and wellness responses through better reasoning, richer context, and clearer communication. The work is backed by physician-informed evaluations aimed at delivering more reliable, trustworthy health guidance.
-
4Using AI to help physicians diagnose rare genetic diseases affecting childrenOpenAI reasoning model aids 18 new rare-disease diagnoses in childrenResearchers used an OpenAI reasoning model to help physicians diagnose rare genetic diseases in children, identifying 18 new diagnoses among previously unsolved cases. The work suggests AI can support complex clinical reasoning and improve diagnosis of rare conditions.
-
5Is it agentic enough? Benchmarking open models on your own toolingHugging Face benchmarks open models' agentic skill on your own toolsHugging Face explores how to judge whether open models are 'agentic enough' by benchmarking them on your own tooling rather than generic suites. The approach evaluates models under realistic, user-specific tool setups to better gauge practical agent capability.
-
6Beyond LoRA: Can you beat the most popular fine-tuning technique?Hugging Face asks if you can beat LoRA, the top fine-tuning methodHugging Face examines whether any approach can beat LoRA, the most popular fine-tuning technique. It compares alternative parameter-efficient methods on performance and cost, probing experimentally whether they can rival LoRA and offering guidance for practitioners.
-
7Anthropic opens Seoul office and announces new partnerships across the Korean AI ecosystemAnthropic opens a Seoul office, announces new Korean AI partnershipsAnthropic opened a Seoul office and announced new partnerships across Korea's AI ecosystem, including enterprises, startups, and researchers building on Claude. It frames Korea as treating innovation and safety as two sides of the same coin. Specifics are per the announcement and unverified independently.
-
8MolmoMotion: Language-guided 3D motion forecastingMolmoMotion: a language-guided approach to 3D motion forecastingAllen Institute for AI (AI2) introduces MolmoMotion on the Hugging Face blog, a method that forecasts 3D motion guided by natural-language instructions. This summary is title-based as no excerpt was retrieved; method details and any performance claims are per the source and unverified independently.