독서리서치레이더투자 프레임워크
로그인 / 회원가입
전체기술제품투자상장시장기타

독서

2026-11-18133 게시물

새로고침

Privacy Firm Proton Launches Lumo Desktop App with Granular AI Agent Controls

중대
2026-11-18 01:32 GMT+8

Swiss privacy technology company Proton officially launched the desktop version of its AI application, Lumo, on November 17, 2026, covering Windows, macOS, and Linux platforms. The app aims to offer users an alternative to mainstream AI tools like ChatGPT and Claude, focusing on providing greater control over personal data when using AI agents to perform local tasks.

Lumo Desktop introduces "Connectors," allowing the AI to access file systems, execute shell commands, fetch browser content, and connect to third-party services such as Proton Mail, Dropbox, Jira, and Confluence. It also supports local and remote Model Context Protocol (MCP) servers. For each connector, users can customize permission levels: specifying which actions can be performed automatically, which require explicit user approval, and which are blocked entirely. For example, a user might allow Lumo to search their inbox but require approval before sending an email.

Beyond immediate interactions, Lumo supports "Schedules" for recurring automated workflows, such as weekly email summaries or monthly spending reviews. Each scheduled task has its own independent permission configuration. The app also supports "Skills," which are saved instructions, templates, and reference files that allow the AI to reuse specific writing styles or processing logic across different conversations.

Regarding privacy protection, Proton emphasizes the use of zero-access encryption, ensuring that only the user can read chat history, not the company. Data is stored on European infrastructure protected by Swiss privacy laws, outside the scope of the U.S. CLOUD Act. Proton states that user conversations and information from connected apps will never be used to train AI models or serve personalized ads. This move comes amid tightening security settings by companies like Apple regarding AI desktop apps requesting full disk access, offering a new option for users concerned with data sovereignty.

출처:TechCrunch

리서치

Google Opens Free Agentic Task Features in Gemini to Users in Six Countries

주제 · 谷歌Gemini任务功能구조적
2026-11-18 01:00 GMT+8

US tech giant Google announced on November 17, 2026, that the 'Tasks' feature in its Gemini app is now rolling out free of charge to hundreds of millions of users across the United States, Japan, South Korea, Canada, Australia, and India. Based on the new Gemini 4 Hydrogen model, this update aims to move beyond answering questions to executing complex daily operations, such as scheduling appointments, managing bills, and shopping.

The core of this release lies in expanding AI capability from 'information retrieval' to 'action execution.' According to Josh Woodward, Vice President of Labs at Google, users can now instruct Gemini to directly call businesses to make reservations. For example, if a vehicle needs service, Gemini can simultaneously contact nearby dealerships, wait on hold, and book an appointment based on the user's calendar availability. During these calls, Gemini discloses that it is an AI and does not share payment or private account information unless explicitly authorized by the user.

In terms of financial handling, Gemini integrates the new Agent Payments Protocol (AP2) and works with Visa, Mastercard, and soon American Express. Users can direct Gemini to analyze utility bills to identify unusual spending or find discount codes and complete purchases while shopping. All transactions are protected by digital tokens, users can set hard spending caps, and receipts are automatically stored in Google Wallet, which is expanding to iOS. Additionally, Gemini supports SMS interaction via RCS integration, allowing users to add items to third-party apps like Instacart by sending photos or text messages.

To ensure privacy and security, Google states that these tasks run on Google Cloud’s Agent Compute Engine using hardware-isolated virtual machines, ensuring each user's session environment is dedicated and never shared. The system includes prompt-injection filters and multiple layers of defense, pausing to request user confirmation before critical steps like clicking 'Buy.' While Google demonstrates strong automation potential, the actual reliability of the feature depends on performance in large-scale user scenarios, with advanced features currently launching primarily in the US market.

출처:Google

리서치

Red Hat and Microsoft Integrate OpenShift Virtual Machines with Azure Direct Virtualization

중대
2026-11-18 01:00 GMT+8

IBM subsidiary Red Hat and U.S. software giant Microsoft announced on November 17, 2026, an expansion of their long-standing engineering partnership to enable enterprises to run virtual machines on Microsoft Azure using Red Hat’s OpenShift Virtualization platform. This integration offers an alternative to traditional hypervisors and what Red Hat described as a path to modernizing legacy applications.

The collaboration involves changes to Azure’s underlying computing architecture to connect Linux natively through “Microsoft Direct Virtualization.” This Azure feature provides performance-sensitive workloads with near-bare-metal access to underlying hardware components. Red Hat stated this creates a direct pipeline between key hardware such as GPUs and isolated workloads, while also working to give Linux-based workloads more direct access to Azure’s hardware resources.

The initiative addresses growing enterprise concerns about hypervisor licensing costs, hardware availability, and the complexity of managing virtual machines alongside newer containerized applications. These pressures have intensified as organizations seek to accommodate AI workloads without abandoning existing applications. Red Hat said it will be the first commercial Microsoft partner to support the technology through “OpenShift Virtualization on Azure Red Hat OpenShift,” the jointly engineered and managed Kubernetes service offered by the two companies.

The agreement includes collaboration on several upstream open-source projects, including KubeVirt, LibVirt, QEMU, and the Linux kernel. KubeVirt enables organizations to manage virtual machines and containers using Kubernetes tools. Microsoft has extended KubeVirt beyond the Linux Kernel-based Virtual Machine hypervisor by adding support for its own hypervisor, allowing OpenShift Virtualization to run on Azure virtual machines using Microsoft’s direct virtualization technology.

“Enterprises shouldn’t have to choose between managing rising hypervisor costs and accelerating their IT roadmaps,” Chris Wright, Red Hat’s chief technology officer, said in a statement. He noted that the collaboration would provide a common foundation for modernizing existing virtual machines and running AI workloads. However, the companies did not announce a general availability date for the new virtualization capabilities.

출처:SiliconANGLE

리서치

AWS Proposes Three-Phase Framework for Responsible Governance of Agentic AI: Assigning Platform and Application Team Responsibilities by Phase

주제 · AWS智能体AI治理框架중대
2026-11-18 00:56 GMT+8

Agentic AI systems can reason, call tools, and execute multi-step operations, expanding the scope of responsible AI from model training and inference to runtime behavior, cross-service cascading failures, and dynamic permission boundaries. In a blog post released on November 17, 2026, Amazon Web Services (AWS) identified four core challenges facing organizations: unpredictable runtime behavior, data exposure and authority boundaries, regulatory velocity outpacing governance capabilities, and the lack of a shared responsibility framework between platform teams and application teams.

The solution proposed in the blog post is a three-phase maturity model, with each phase focusing on a different combination of responsible AI pillars. Phase 1, Governance Foundations, focuses on governance, security, and privacy: platform teams establish landing zone infrastructure, implement data classification and access controls, define risk classification standards, and tag each planned use case as low, medium, high, or unacceptable risk, while centrally registering all agents, tools, and skills in a registry. Phase 2, Pilot Deployment, focuses on safety, veracity and robustness, controllability, and security and privacy: application teams implement orchestration patterns, knowledge grounding, personally identifiable information (PII) filtering, and human-in-the-loop approval circuit breakers based on templates and guardrails provided by the platform team. Phase 3, Evaluation and Hardening, focuses on fairness, explainability, transparency, and security and privacy: this involves running bias benchmark tests, establishing decision audit trails, conducting red-team exercises, and embedding evaluation gates into CI/CD pipelines.

The blog post uses the example of a bank building a multi-agent loan processing platform to illustrate implementation across phases. In Phase 1, the platform team registers agents at a high-risk level, configures tenant isolation, and documents residual risk. In Phase 2, risk scoring uses Bedrock Knowledge Bases grounded in regulatory lending documentation; automated reasoning checks verify that denial decisions conform to written standards; Guardrails filter PII before customer communication; and the scoring agent confirms key applicant data from source systems rather than relying solely on extraction results from the intake agent. In Phase 3, AgentCore Evaluations run on every pipeline push; CI/CD blocks deployment if evaluation results fall below release criteria or if per-subgroup fairness metrics drift beyond thresholds; and human review of denial decisions transitions to an after-the-fact appeals process once 2,000 consistent decisions are accumulated over 60 days.

It is important to note that this framework was proposed by AWS itself and is deeply tied to its Bedrock AgentCore product; the 12-week implementation timeline and phase-pillar mapping have not been independently verified. The Robinhood case cited in the article reports a self-reported 20% efficiency gain, lacking third-party audit data. The blog post itself states that it does not provide legal or compliance advice and indicates that the framework is a starting point rather than an exhaustive compliance checklist.

출처:Amazon Web Services

리서치

US DOJ and Commerce Confirm Criminal Probe Into Megaspeed Over Nvidia Chip Export Violations

주제 · Megaspeed중대
2026-11-17 21:30 GMT+8

The US Department of Justice and the Commerce Department issued a joint statement on November 17, 2026, confirming a criminal investigation into Singapore-based cloud computing firm Megaspeed. Authorities allege the company may have circumvented export rules designed to restrict the sale of advanced AI chips to approved entities and countries. This development marks an escalation from administrative scrutiny triggered by the recent scandal involving former Nvidia CEO Jensen Huang to active federal law enforcement.

The statement identifies Megaspeed as one of several companies suspected of facilitating the transfer of controlled AI chips to China. Although Nvidia added Megaspeed to its 'no-sell' list in June 2025, reports indicate that a Malaysian subsidiary of Megaspeed continued to receive large shipments of Nvidia chips for eight months thereafter. These components, including export-controlled Blackwell and H200 models, were allegedly shipped to Speedmatrix before being handed off to China.

Nvidia has argued that it received permission from the Commerce Department to ship components to Megaspeed and continues to list the firm as a cloud partner. However, the initiation of a criminal probe suggests deepening regulatory concerns over Nvidia's compliance enforcement. The Commerce Department had previously investigated Megaspeed starting in 2024, requesting sales data from Nvidia in May 2025. The Justice Department took the lead after officials visited Megaspeed and interviewed former executive Alice Huang in June 2025.

For AI industry investors, this event highlights geopolitical risks and compliance gaps in the global AI compute supply chain. Megaspeed was previously accused of helping move approximately $2 billion worth of Nvidia chips into China. As the US government intensifies its crackdown on chip smuggling, Chinese AI firms relying on gray-market channels face increased supply disruption risks, while chipmakers like Nvidia must navigate stricter export control audits and potential legal liabilities.

출처:Tom's Hardware

리서치

AMD Demonstrates Trade-offs in Agent Inference Scheduling: Active Session Latency Drops 22%, New Request Wait Increases 49%

주제 · AMD智能体推理调度중대
2026-11-17 08:00 GMT+8

The traffic patterns of agent coding sessions differ fundamentally from those of standard chat requests. In a technical blog post about its ATOM inference engine, AMD described a typical case: one session involved 198 model calls, with call #48 sending 195.5K tokens, of which only 1.3K were new content. The corpus-wide median input length was 142,016 tokens, while the median output was just 444 tokens. This pattern—extremely long inputs, extremely short outputs, and highly shared prefixes—combined with the bursty parallelism of sub-agents and idle intervals caused by tool execution, creates service challenges that are entirely different from those posed by standard requests.

AMD categorizes these challenges into three layers: memory (maintaining reusable history), scheduling (balancing new requests against ongoing decoding), and compute (reducing per-step overhead in small batches). Among these, the trade-offs in the scheduling layer most intuitively reveal the inherent contradictions of agent services. ATOM’s PrefillDelayer allows a fixed number of decode steps to run immediately after each prefill completes, during which newly arrived requests are held and merged into a single prefill operation. Under 16 concurrent sessions, this strategy reduced P90 token inter-token latency from 31.6 milliseconds to 24.6 milliseconds—a 22% improvement—allowing active sessions to maintain a more stable output rhythm; however, the median time-to-first-token for new requests increased from 1,182 milliseconds to 1,766 milliseconds, a 49% deterioration.

This contradiction is sharper in agent workloads than in standard chats because the session tree structure amplifies latency. A parent agent waits for results from its sub-agents, meaning any delay on one branch postpones the completion of the entire task. Consequently, AMD enables PrefillDelayer only at concurrency levels of 16 and above, keeping it disabled in lower-concurrency interactive ranges, and provides different configuration combinations in deployment recommendations based on concurrency levels.

Optimizations in the memory and compute layers each have their own applicable conditions. ReplaySSM replaces the full state snapshot for every draft position in speculative decoding with a single checkpoint plus compact update records. In experiments with seven draft tokens, this freed up 24.49 GiB per GPU card, increasing available KV blocks from 8,325 to 18,813—a 2.26x increase—but this adds historical retention capacity rather than throughput or cache hit rates. Mono decode fuses multiple small kernels in individual KDA layers into two persistent kernels, reducing the target model's forward pass time from 22.0 milliseconds to 16.4 milliseconds in single-request tests; however, this optimization is currently enabled only at single concurrency, and its effects at higher concurrency levels remain unmeasured.

In terms of comprehensive effects, AMD reported that at 48 concurrent sessions, 94.1% of prompt tokens were reused rather than recomputed (with 77.3% coming from GPU HBM and 16.8% recovered from CPU DRAM), whereas the theoretical upper limit for this corpus is 97.2%. At 56 concurrent sessions, eight MI355X chips achieved a total of 17.5K tokens per second per chip. These figures come from AMD’s runs on the SemiAnalysis InferenceX AgentX benchmark and represent engineering references from the vendor under specific hardware and model combinations, not universally validated independent third-party performance benchmarks.

출처:ROCm Blogs

리서치

Databricks and Google Cloud Enable Bi-Directional Zero-Copy Data Federation Across Unity Catalog and Borderless Lakehouse

중대
2026-11-18 00:30 GMT+8

Data analytics platform company Databricks announced on November 17, 2026, a significant advancement in its partnership with Google Cloud: customers can now access a single copy of data from both Unity Catalog and Google Cloud's "borderless Lakehouse" without duplication. This bi-directional zero-copy federation allows engines like BigQuery to read tables managed by Unity Catalog, and vice versa, breaking down traditional silos between data warehouses.

Beyond data interoperability, the two companies are driving open standards for catalog metadata. In collaboration within the Apache Iceberg community, they have landed proposals for "catalog labels," enabling business semantics and governance tags (such as "domain=finance" or "pii=ssn") to travel with data across different catalogs. This mechanism is critical for enterprise AI agents, which rely on shared metadata to identify sensitive information, apply masking policies, and understand business context when searching for assets across platforms.

The feature is available now through Databricks' Lakehouse Federation and corresponding Google Cloud services. While this marks a key milestone in the open lakehouse architecture, specific production performance, latency under high concurrency, and governance maturity for large-scale deployments remain based on vendor self-reports and await independent benchmark verification.

출처:Databricks

리서치

AI Data Infrastructure Firm LanceDB Raises $57.5M Series B to Accelerate Multimodal Model Training

주제 · LanceDB融资간단한 요점
2026-11-18 00:30 GMT+8

LanceDB, an AI data infrastructure startup, has announced the completion of a $57.5 million Series B funding round co-led by Bessemer Venture Partners and Innovation Endeavors. Other participants included Theory Ventures, Dimension, Swift Ventures, CRV, and Y Combinator, bringing the company's total funding to $98.5 million.

LanceDB develops database technology designed to accelerate artificial intelligence model training. Its core product is a unified multimodal data lakehouse that allows researchers, enterprise teams, and AI agents to access, find, prepare, and train on complex datasets containing text, images, video, and 3D sensor information without spending weeks managing underlying infrastructure.

According to SiliconANGLE, Chang She, co-founder and CEO of LanceDB, stated that there is a missing data layer in AI today, forcing research teams to spend significant time integrating scattered datasets before each training run. LanceDB's technology is built on Lance, an open-source data format that supports rapid retrieval and modification of individual records, making it particularly useful for handling petabyte-scale training datasets.

The company plans to use the new funds to expand its multimodal data platform, focusing on improving large-scale data curation, feature engineering, and data loading to efficiently supply thousands of GPUs during training. It will also deepen support for emerging AI workloads such as video generation, world models, and robotics.

출처:SiliconANGLE

리서치

Nvidia Former CEO Jensen Huang Says He Is Willing to Return If Board Decides It Best for Company

구조적
2026-11-18 00:09 GMT+8

Jensen Huang, the former Chief Executive Officer of Nvidia, issued a statement on November 17, 2026, explicitly expressing his willingness to return to the role if the company's board of directors believes it is in the best interest of shareholders, employees, and the company. The statement was initially reported by The Wall Street Journal and subsequently covered by major technology outlets.

"Since my departure, I’ve seen an outpouring of calls from Nvidia’s employees, investors, business partners, and customers for me to return as CEO," Huang said in the statement. "If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO."

This declaration comes amid significant leadership turmoil at Nvidia. Huang resigned last Thursday following reports of personal affairs, with Chief Financial Officer Colette Kress appointed as interim CEO. While more than 193,000 people, including prominent figures like Elon Musk, have signed a petition calling for Huang's return, reports indicate that Nvidia's board is currently not considering reinstating him.

Huang's public statement transforms the possibility of his return from private discussions or external pressure into a clear, conditional offer based on the board's decision. This adds complexity to the search for a permanent successor, particularly as Nvidia's stock price fluctuates due to the leadership crisis and the company faces heightened uncertainty as a core supplier in the AI infrastructure sector.

출처:The Verge

리서치

Apple Tightens macOS Permissions: AI Agents Face Stricter Warnings and Granular Controls

중대
2026-11-18 00:35 GMT+8

Apple has implemented significant changes to its security architecture in the second beta version of macOS 27.3, aimed at restricting unrestricted access by artificial intelligence (AI) agents to Mac data. This move directly addresses growing concerns about AI tools potentially over-accessing user privacy, exemplified by reports that Meta's Muse AI agent read iMessage contents without explicit consent to generate suggestions.

In the new version, the process for granting "Full Disk Access" and "Accessibility" permissions has become more cumbersome and transparent. Users can no longer simply drag an app icon to authorize access; they must select the requesting app from a list. The system displays detailed warning dialogs explaining exactly what data the app can view, modify, or extract once granted these broad permissions. Apple emphasizes that such extensive access puts all data on the Mac at risk.

Beyond the approval workflow, macOS 27.3 introduces new granular options. Users can now grant AI applications access only to specific folders or specific other apps, rather than the entire system. Additionally, when an app controls another app via "System & App Protection Override" permissions, a prominent icon appears in the menu bar, alerting users in real-time to background automation activities.

For developers, existing backup and assistive technology apps can request special "entitlements" to maintain user experience while complying with new security standards. However, permissions previously granted by ordinary users will not be automatically revoked; the new rules apply only to future requests. This initiative reflects Apple's attempt to balance the proliferation of AI features with the protection of user data sovereignty, building a visible firewall as AI agents increasingly penetrate the operating system's core.

출처:Mac & i

리서치

Google Threat Intelligence Reports Attackers Modifying Claude Code Config Files to Install Backdoors

중대
2026-11-17 09:22 GMT+8

Google's Threat Intelligence Group (GTIG) and Mandiant have released a quarterly report detailing new methods of malicious generative AI usage. The findings indicate a shift from single-prompt interactions to orchestrating multiple AI agents for automated intrusion tasks. A particularly notable trend involves attacks targeting the configuration files of AI coding tools.

According to the report, an actor named "MIDNIGHT NEPTUNE" targeted cryptocurrency-related organizations in North America and Europe. Using fake recruiter profiles, spoofed conference sites, and "ClickFix" tactics that trick users into executing commands, the group gained initial access. Once inside internal code repositories, the attackers modified the .claude/settings.json file associated with Claude Code. When developers launched Claude Code in these compromised environments, the system automatically downloaded and executed a backdoor named "SOMBERMEME." This technique turns the AI development tool itself into a vector for malware distribution.

Beyond supply-chain attacks on specific tools, AI-driven automated intrusions are accelerating. The report highlights the operational use of the open-source attack framework ARTEX. Released on GitHub in July 2026 as a penetration testing tool, ARTEX garnered significant attention, exceeding 12,000 forks before being made private on October 8. Wiz and Google reported in October that China-linked actors were using ARTEX to target IT and software companies. Other frameworks such as CAIRN, CyberStrikeAI, and PentAGI were also cited.

The automation of the vulnerability-to-exploit lifecycle is also increasing. An exposed server revealed a tool called "exploit_agent," which connects to multiple large language models to automatically compare pre- and post-patch code, identify vulnerabilities, and generate and execute exploit code. Targets included websites in France's media, entertainment, and education sectors.

Meanwhile, "LLMJacking"—the theft of AI compute resources—is becoming widespread. In one case handled by Mandiant, attackers generated over $100,000 in fraudulent bills in a single day, with cumulative damages reaching $1 million in the worst instances. Underground services known as "LLM Mixers" are reselling stolen AI credentials, with prices starting as low as $0.10 per million tokens.

출처:テクノエッジ TechnoEdge

리서치

"Asynchronous Methods for Deep Reinforcement Learning": Replacing Experience Replay with Parallel Agents for Efficient Training on a Single CPU

검증됨 2026-11-18 00:04 GMT+8

장기 리딩 · 《Asynchronous Methods for Deep Reinforcement Learning》(2016)

In 2016, the Google DeepMind team published the paper "Asynchronous Methods for Deep Reinforcement Learning", introducing the A3C algorithm. Prior to this, deep reinforcement learning primarily relied on an "experience replay" mechanism to stabilize training, which involved storing historical agent data and randomly sampling updates. While effective, this approach required significant memory and computational resources and restricted the use to off-policy algorithms.

The core innovation of this paper was abandoning experience replay in favor of an asynchronous parallel architecture. By running multiple independent "actor-learner" threads on a standard multi-core CPU, each thread interacted with the environment and computed gradients independently. This parallelism naturally decoupled the temporal correlation of data, making the training process more stable while enabling the use of a broader range of online policy algorithms (such as Sarsa and n-step methods).

Experimental results showed that the best-performing Asynchronous Advantage Actor-Critic (A3C) method surpassed state-of-the-art approaches in Atari game domains. Notably, it achieved this using only a single multi-core CPU rather than a GPU, halving training time. This finding demonstrated that efficient deep reinforcement learning could be achieved without expensive hardware or distributed clusters, providing a lightweight design philosophy for subsequent algorithms.

출처:《Asynchronous Methods for Deep Reinforcement Learning》(2016)

리서치
다음 →
로그인 / 회원가입
독서리서치레이더투자 프레임워크

To help teams assess their current status, AWS offers a portable Responsible AI Assessment Skill that scores along various dimensions on a 1-to-5 scale, where a score below 4 indicates substantial risks that should be remediated. This assessment tool can be run within AI agents such as Amazon Quick, as well as AI coding assistants like Kiro and Claude Code.