2026-06-19 AI 日报

The US says ASML’s top chip tool may be in China, but how?

MosaicLeaks: Can your research agent keep a secret?

Back to Articles TL;DR Deep research agents increasingly combine private local documents with external tools like web retrieval, creating a privacy risk: an agent's external queries may leak sensitive information. MosaicLeaks proposes a new

Hugging Face · 阅读原文

LedgerAgent: Structured State for Policy-Adherent Tool-Calling Agents

Policy-adherent tool-calling agents in customer-service domains must maintain task states across turns while calling tools and obeying domain policies. Task states consist of relevant facts, identifiers, constraints, and conditions observed through user interaction and tool calls. In standard agents, task states are not represented separately. Observations, tool returns, and policy instructions are placed in the prompt, leaving agents to reconstruct the relevant states from the prompt each time they decide what to do next. This design makes state management implicit, creating two common failure modes. An agent may retrieve the right facts but later ground its decision in stale, missing, or incorrect information; and a syntactically valid tool call may still violate a domain policy that depends on the current task state. We introduce \textsc{LedgerAgent}, an inference-time method for tool-calling agents that maintains observed task states in a separate ledger and renders the states into the prompt. The ledger is also used to check state-dependent policy constraints before environment-changing tool calls are executed, blocking policy violations. Across four customer-service domains and a mixed panel of open- and closed-weight models, \textsc{LedgerAgent} improves average pass\textasciicircum{}k over a standard prompt-based tool-calling approach, with the largest gains under stricter multi-trial consistency metrics.

‘Queer Eye’ life coach Karamo Brown launches Kē, a wellness app featuring his AI digital clone

《粉雄救兵》主持人 Karamo Brown 推出健康应用 Kē，上线 iOS 和 Android，订阅价为每月 14.99 美元。应用集成个性化健身计划、基于家中食材的营养建议、冥想内容和社群功能，并允许用户通过 AI 聊天机器人调整训练与饮食方案。其差异化卖点是“AI Karamo”数字分身：基于 Delphi 技术，利用 Brown 的采访、播客等素材生成可实时语音互动的虚拟顾问。报道还提到，该产品设置了人工审核与安全防护，但用户与 AI 的对话数据会分享给 Delphi，未来 Delphi 还计划为 Kē 增加可代用户执行任务的 agent 能力。

行业

More people get news from AI chatbots, but trust remains low

The Decoder · 阅读原文

The CEO of Allbirds’ new AI biz has a plan, but no team

Source: Elastic agrees to buy CRV-backed Deductive AI for up to $85M

Deductive AI, a startup that uses AI to catch and resolve bugs in software, has agreed to be sold to enterprise software company Elastic for up to $85 million, according to a person with knowledge of the deal. Deductive, which was founded i

AI inference startup Baseten reportedly raising $1.5B months after its last mega-round

In Brief Posted: 2:20 PM PDT · June 18, 2026 Image Credits:Nuthawut Somsuk / Getty Images AI inference company Baseten is close to finalizing a stunning $1.5 billion funding round at a $13 billion valuation, the Wall Street Journal reports.

Snap spins off AI video team into new company, Dotmo, due to costs

Snap will be spinning off an internal generative AI video team into a separate company. The new company — dubbed Dotmo — will focus on developing AI models that can create interactive gaming experiences, Snap told TechCrunch. Snap cited the

OpenAI is bringing on some big guns in the lead-up to its IPO

OpenAI is bringing on some big names to the team in the lead-up to its public debut: Google DeepMind AI legend Noam Shazeer and former Trump White House AI policy official Dean Ball. Shazeer, a co-lead at Gemini and the founder of AI role-p

Almost half of US singles feel negatively about AI in dating, Match says

Dating app giant Match Group — which owns apps like Tinder, Hinge, and OkCupid — conducted a study to determine how U.S. singles really feel about the relationship between AI and dating. Turns out, people don’t want AI messing with every as

Amazon hopes to challenge Nvidia more directly by selling its AI chips

据报道，AWS 正在洽谈将自研 AI 芯片 Trainium 销售给其他公司用于数据中心，这意味着亚马逊可能从仅在云内部消化芯片，转向更直接挑战英伟达的对外供货模式。亚马逊 CEO Andy Jassy 此前称，若芯片业务独立并同时向 AWS 与第三方销售，年化规模可达约 500 亿美元。不过报道也指出，此事仍处早期阶段，且 AWS 当前与下一代 Trainium 产能都已非常紧张，对外销售将受制于制造与供应能力。

Structuring and Tokenizing Distributed User Interest Context for Generative Recommendation

Generative recommendation is an emerging paradigm that has shown promise in industrial recommendation systems, aiming to predict users' next interactions from their historical behaviors. At the core of generative recommendation lies item tokenization, which bridges item semantics and recommendation models. However, existing methods often struggle to effectively organize and inject complex user-behavioral and item-semantic contexts into recommendation models simultaneously. On the one hand, existing graph-based integration methods, such as graph serialization and graph neural networks, either suffer from scalability issues or exploit only local graph information. On the other hand, existing semantic tokenization methods typically rely on heuristics and lack explicit supervision signals, which may lead to inaccurate or suboptimal semantic representations. To address these limitations in user interest context modeling, we propose G2Rec, a scalable framework that unifies holistic graph-based user co-engagement modeling with semantic tokenization for industrial-scale generative recommendation. Overall, G2Rec enables recommendation models to capture holistic and semantically grounded user interest prototypes without requiring ground-truth user interests, thereby providing more comprehensive and accurate modeling of user behavior contexts in industrial sequential recommendation. Online deployment across product surfaces and extensive experiments on public datasets demonstrate the superiority of G2Rec over existing methods.

Toward Calibrated Mixture-of-Experts Under Distribution Shift

Calibration aligns a model's predictive uncertainty with the frequencies of its empirical outcomes and is important for understanding and trusting reported probabilities. Recent work shows that enforcing calibration at the level of individual predictors can improve ensemble accuracy and calibration, with mixture-of-experts (MoE) models showing strong empirical improvements in particular; however, the conditions under which calibration helps MoE are not well understood. In this work, we study how MoE models behave under distribution shift, focusing on how routing mechanisms interact with expert-level calibration. We show that expert calibration is sufficient to ensure calibration of the overall model under a broad class of distribution shifts in hard-routed models, but is insufficient for calibrating soft-routed models. To address this, we propose an adversarial reweighting that penalizes calibration errors of the routed aggregate under distribution shift, and we demonstrate that it improves the accuracy-calibration tradeoff both on average and on difficult subsets of the data, across model classes, prediction tasks, and distribution shifts.

AI data centers just got a government-mandated fast lane to the grid

The Federal Energy Regulatory Commission (FERC) told grid operators on Thursday to fast-track interconnection requests from data centers and other large electricity users. Under the orders, six major grid operators have to show that data ce

How Do Instructions Shape Speech? Cross-Attention Attribution for Style-Captioned Text-to-Speech

Style-captioned text-to-speech systems use natural language to control voice characteristics, but how individual words influence acoustic output remains unclear. Understanding this is critical for diagnosing failure modes and improving controllability in expressive TTS. We propose cross-attention attribution for speech diffusion models, adapting the DAAM framework to the speech domain for the first time, and apply it to CapSpeech-TTS. Our method extracts per-token heatmaps across 25 layers and 24 ODE steps. We analyze 3,600 (style caption, text transcript) combinations comprising 120 style captions conditioning the generation of 30 text transcripts each, revealing how caption tokens shape waveforms. Results show: (1) style tokens have lower temporal variance than content/function tokens, confirming global conditioning; (2) style attention correlates with F0 and energy; (3) style conditioning peaks in early steps and deep layers; (4) attention entropy reaches its minimum at layer 17, co-occurring with the style importance peak, indicating maximal network selectivity at the most style-critical stage. This is the first study of how natural language influences cross-attention in speech diffusion models

StylisticBias: A Few Human Visual Cues Drive Most Social Biases in MLLMs

Multimodal large language models (MLLMs) are increasingly deployed in personally and societally consequential settings, yet the visual cues that shape how these models judge people remain poorly understood. Prior work often compares different (groups of) individuals, making it difficult to separate appearance effects from identity differences. We introduce StylisticBias, a controlled benchmark for evaluating attribute-level social bias in MLLMs. We generate 500 photorealistic base faces and create about 50 single-attribute variations per face, producing about 25K images. This design keeps identity fixed and changes one visual attribute at a time. It lets us measure how specific cues shift model judgments. We evaluate six MLLMs across 25 binary social judgment scenarios. We find that age and body type dominate identity-level effects, while fashion style and other visual cues drive the largest attribute-level shifts. We further find that about 15 attributes account for nearly 80\% of the total variation, showing that bias is concentrated in a small set of visual cues. Sensitivity is strongest in judgments that are semantically aligned with appearance, especially socioeconomic and style-related judgments. We release StylisticBias as a benchmark for fine-grained bias evaluation in multimodal models. Code and dataset: https://github.com/timo-cavelius/StylisticBias and https://hf.co/datasets/shaghayegh/stylistic-bias-dataset.

DeepSWIP: Quotient-WMC Counterfactuals for Neural Probabilistic Logic Programs

Neurosymbolic systems such as DeepProbLog combine neural perception with probabilistic logic, but standard inference is associational. Counterfactual reasoning additionally requires a causal semantics for interventions and evidence. We introduce DeepSWIP, a single-world counterfactual semantics for DeepProbLog programs. Using neural materialization, we reduce fixed-context neural predicates to ordinary ProbLog choices, apply Single World Intervention Programs (SWIPs), and compute counterfactuals by weighted model counting (WMC) over a single transformed program. Under finite grounding and unique-supported-model assumptions, DeepSWIP is exact relative to the learned materialized FCM. The standard quotient-WMC form of ProbLog conditionals identifies active neural probabilities and explains intervention cleaning, calibration sensitivity, and rare-evidence instability. Experiments on MPI3D confirm the transformation against a DeepTwin construction against 12,000 queries, as predicted and a 2.14$\times$ inference speedup from avoiding the Twin's endogenous duplication. A SUMO HOV experiment shows that neural calibration degradation biases plug-in estimates, while a correctly scoped randomized-policy AIPW estimator removes most first-order bias for population mean and ATE estimands. Code is at https://github.com/saibib/deep_SWIP.

SARLO-80: Worldwide Slant SAR Language Optic Dataset 80cm

Multimodal foundation models have advanced rapidly thanks to large optical benchmarks, but comparable resources for synthetic aperture radar (SAR) remain limited. Existing SAR--optical datasets largely rely on low-resolution, intensity-only Ground Range Detected~(GRD) products and do not preserve complex-valued SAR measurements or native acquisition geometry, which restricts physically grounded multimodal learning. In particular, large-scale public datasets combining very-high-resolution (VHR) SAR SLC, aligned optical imagery, and natural-language descriptions are still lacking. We present a VHR SAR--optical--text dataset built from open-access Umbra spotlight acquisitions distributed as Sensor Independent Complex Data (SICD). From around 2,500 worldwide scenes (VV/HH, 20cm--2m native resolution), we standardize all SAR data to an 80cm slant-range grid via band-limited FFT resampling and tile the imagery into 1024 by 1024 patches. For each SAR patch, we retrieve a high-resolution optical tile and warp it into the SAR grid using local coordinate correspondences for local pixel-level alignment. We further generate three caption variants (SHORT/MID/LONG) per sample to support vision--language training and evaluation. Our dataset contains 119,566 triplets (complex and amplitude slant-range SAR patch, aligned optical patch, natural-language description) covering 257 locations across 72 countries and a broad range of land types and infrastructures. We release fixed train/validation/test splits and the full preprocessing and baseline code to enable reproducible benchmarks for multimodal alignment on cross-modal retrieval and conditional generation in native SAR geometry. The dataset is publicly available on the Hugging Face Hub at https://huggingface.co/datasets/ONERA/SARLO-80.

Sovereign Execution Brokers: Enforcing Certificate-Bound Authority in Agentic Control Planes

Autonomous agents are increasingly connected to cloud, deployment, and data-control workflows, but production mutation authority should not reside inside non-deterministic reasoning processes. Existing access-control mechanisms authorize identities, while assurance layers certify proposed actions; neither alone provides a mandatory enforcement point for certified authority at the moment of mutation. This paper introduces the Sovereign Execution Broker (SEB), a runtime enforcement boundary for certificate-bound agentic infrastructure. SEB consumes certificates issued by the Sovereign Assurance Boundary (SAB), verifies that the requested mutation matches the certified execution contract, checks validity windows, policy epochs, revocation epochs, and live-state drift, mints scoped execution identity, invokes infrastructure APIs, and records signed decision and outcome records. By separating proposal, admission, and execution, SEB turns certified authority into a short-lived, revocable, auditable runtime capability, provided that production mutation APIs reject non-broker identities. We present the SEB execution model, certificate and replay-verification predicates, scoped identity semantics, bypass-prevention deployment patterns, failure behavior, and a concrete prototype implementation. We evaluate the prototype on AWS and Kubernetes clusters, measuring latency overheads, revocation propagation, drift detection, and security under fault injection.

The smartphone era created an attention crisis — slow tech is fixing it

When Tony Fadell entered New York City’s 28th Street Subway Station, he did not expect to come face-to-face with an advertisement for a product he designed over 20 years ago. But there it was: a five-by-four-foot poster promoting the iPod S

New usage analytics and updated spend controls for enterprises

OpenAI introduces new spend controls and usage analytics for ChatGPT Enterprise, helping organizations manage costs and scale AI with confidence.

OpenAI · 阅读原文

论文

How Transparent is DiffusionGemma?

这篇论文研究扩散式语言模型 DiffusionGemma 的推理透明性，并将其拆分为“变量透明性”和“算法透明性”两部分。作者指出，若按直观方式衡量，DiffusionGemma 在可解释状态之间的“不透明串行深度”约为自回归 Gemma 4 的 28.6 倍；但如果把去噪步骤之间的信息映射到可解释的 token 瓶颈，且不损失下游性能，则这一差距可降到 1.1 倍。论文还通过案例分析总结了扩散模型特有的解释学现象，包括非按时间顺序推理、token/序列 smearing，以及中间上下文推理。最后作者测试了可监控性，结果显示 DiffusionGemma 与 Gemma 4 大致相当。对关注可解释性、安全监控和扩散式 LLM 的研究者有较高参考价值。

FlowEdit: Associative Memory for Lifelong Pronunciation Adaptation in Flow-Matching TTS

这篇论文提出 FlowEdit，用于解决流匹配文本转语音（Flow-Matching TTS）模型上线后对生僻专有名词发音错误难以修复的问题。其核心做法是不更新模型权重，而是在文本嵌入空间学习 token 级扰动作为发音修正，并将修正结果存入现代 Hopfield Network 作为可检索的联想记忆；推理时通过软注意力和相似度门控召回历史修正，支持模糊形态匹配。论文在覆盖18个语系、312个多语言专有名词的基准上，目标词音素错误率相对零样本基线下降92.7%，同时保持通用语音质量不变，单次修正约需15秒单卡GPU。对中文AI从业者而言，这对TTS产品中的持续学习、低成本热修复发音错误和个性化词典适配有较强参考价值。

技巧

How to turn off AI in your Google Docs

这是一篇面向普通用户的实用教程，介绍如何在 Google Docs 中关闭 Gemini 的“write with Gemini”等 AI 提示。文中给出两种方式：一是在 Google Docs 顶部菜单进入 Gemini 的“bottom bar preferences”关闭底部栏；二是到 Gmail 设置中进入“Google Workspace smart features”，关闭 Workspace 智能功能，从而一并减少 Google Docs 中的 Gemini 弹窗与写作辅助提示。

来源引用

RT Derya Unutmaz, MD: I’m very excited about this article from @OpenAI on my attempt to use GPT-5 Pro to understand the results of an experiment we d...X：OpenAI (@OpenAI) · 原始资料与分析线索

Anthropic is donating another $20 million to Public First ActionAnthropic · 原始资料与分析线索

美国政府下令暂停 Anthropic Fable 5 与 Mythos 5 的全球访问Anthropic · 原始资料与分析线索

Introducing Claude for TeachersAnthropic · 原始资料与分析线索

More details on Fable 5’s cyber safeguards and our jailbreak frameworkAnthropic · 原始资料与分析线索

Expanding Project GlasswingWe’re extending Project Glasswing to approximately 150 new organizations in more than fifteen countries.Anthropic · 原始资料与分析线索