LesenForschungRadarAnlageframework
Anmelden / Registrieren
AlleTechnikProdukteInvestierenBörseSonstiges

Lesen

2026-11-0499 Beiträge

Aktualisieren

Vercel Launches Public Beta of Security Review: AI Automatically Audits Every Code Push

Thema · Vercel安全审查功能Wesentlich
Verifiziert 2026-11-04 17:50 GMT+8

US cloud development platform Vercel has announced that its "Security Review" feature is now in public beta for Pro and Enterprise teams. The tool aims to integrate AI-driven security analysis directly into the daily code submission process, rather than serving only as an after-the-fact audit or an add-on check for pull requests.

Unlike traditional scans based on pull requests, Security Review is triggered by push events to connected GitHub repositories. This means that both regular commits and direct force pushes to production branches undergo immediate automated vulnerability review. Each finding records the project, environment, branch, and commit, accompanied by an AI-generated severity rating and a suggested fix.

In the remediation phase, developers can click "Fix with Vercel Agent," allowing Vercel's AI agent to propose a fix and open a pull request for human review, ensuring nothing merges without approval. Additionally, teams can write security policies in plain language to provide context for the AI review.

Vercel explicitly notes the limitations: the tool analyzes source code only, does not perform runtime or penetration testing, and therefore cannot replace a full security audit nor guarantee an application is free of vulnerabilities. Pricing is billed at the Vercel Agent rate, including underlying model inference costs plus a Vercel Token Rate of $0.25 per million tokens. Reviews are free until November 10, 2026, but subsequent usage will incur costs scaling with push frequency and code change volume.

Quellen:Vercel

Forschung

Vercel Launches Security Review Beta: AI Agents Scan and Fix Code Vulnerabilities on Push

Thema · Vercel安全审查功能Wesentlich
2026-11-02 21:00 GMT+8

US cloud development platform Vercel announced on November 2, 2026, the launch of "Security Review," a new capability within its AI agent product, Vercel Agent. Currently in public beta, the feature aims to address the growing bottleneck of code security reviews in the era of AI-assisted programming by integrating automated scanning and remediation into every code push.

The mechanism is triggered by GitHub push events. When developers submit code to repositories connected to Vercel projects, the AI agent immediately scans the changed parts (diffs) for vulnerabilities such as missing authorization checks or exposed secrets. Unlike traditional pattern-matching scanners, Vercel claims its agent reads surrounding context, traces code reachability, and identifies logic flaws. Findings appear as comments on Pull Requests and are aggregated in the Security dashboard.

Beyond detection, the feature emphasizes remediation. Users can click "Fix with Vercel Agent" in the dashboard, prompting the AI to verify the vulnerability, generate a minimal fix, and update tests. Teams can also define security policies in plain language (e.g., requiring all endpoints returning customer data to verify requester ownership), which are applied during each review. For enterprise users, each new finding generates an audit log event sendable to SIEM systems.

Vercel notes that as frontier AI models make it easier to find and exploit vulnerabilities, and as AI agents write more production code, traditional security review methods struggle to keep pace. Security Review currently supports code pushed to GitHub, with plans to extend to Vercel CLI deployments, v0, and non-Vercel hosted repositories. The feature is available for Pro and Enterprise teams, billed at the Vercel Agent rate, but free until November 10, 2026.

Quellen:Vercel

Forschung

Chunghwa Telecom Integrates Google AI Subscriptions into Telecom Billing, Targets 200k Paid Users in One Year

Wesentlich
2026-11-04 16:52 GMT+8

Chunghwa Telecom announced on November 4, 2026, an expanded partnership with Google to integrate Google AI Plus and Google AI Pro plans into its telecommunications service system. This means consumers in Taiwan can now subscribe to these services through Chunghwa Telecom, including bundling them with 5G mobile or HiNet broadband plans, in addition to subscribing directly from Google.

The key difference lies in the sales channel and payment model, not in AI capabilities. Regardless of the acquisition method, users still access Google's Gemini AI features and corresponding Google One cloud storage. Chunghwa Telecom has not launched different versions of AI services but rather serves as a new sales entry point for Google's AI subscriptions.

Unlike Google's official annual plan which requires upfront payment for the full year, Chunghwa Telecom's offering allows users to pay monthly under an annual contract, lowering the initial barrier for consumers trying paid AI services. Additionally, existing Chunghwa Telecom customers can acquire AI services through their familiar phone or broadband bills, potentially reducing costs further when bundled with specific plans. This model mirrors how telecom operators previously integrated streaming or cloud services, making AI part of value-added telecom services.

As of September 2026, Chunghwa Telecom had over 13 million mobile users and 4.48 million broadband users. The company aims to achieve more than 200,000 paid Google AI (Gemini) users within one year through this collaboration. 林雅芳, General Manager of Google Taiwan, stated that future AI popularization relies on cooperation with local partners, hoping to leverage Chunghwa Telecom's extensive distribution network to bring Google AI services into daily life.

Quellen:iThome

Forschung

Microsoft .NET Team Experiments with Packaging AI Agent Skills in NuGet

Thema · NuGet代理技能分发Wesentlich
2026-11-04 14:18 GMT+8

The Microsoft .NET team has published experimental results demonstrating how to encapsulate AI Agent Skills within NuGet packages. This approach allows library developers to provide corresponding usage guidelines when releasing new versions, helping AI programming agents better understand how to use new APIs and reducing errors caused by reliance on outdated model training data.

Agent Skills is a mechanism that provides operational instructions to AI agents via SKILL.md files, which document applicable scenarios, execution steps, and related knowledge. Previously, these skills were primarily distributed through Git repositories or skill marketplaces. Microsoft's experiment uses NuGet as a carrier, enabling different versions of libraries to carry their own applicable documentation. When a .NET project executes dotnet restore, NuGet downloads dependencies to the local cache, including any embedded SKILL.md files.

However, downloading a package does not mean the AI agent will automatically use the skills contained within it. Agents must still discover and read these files themselves, or use accompanying tools to find and import them. Microsoft showcased two auxiliary tools: dotnet-inspect, which checks for skills included in NuGet packages or project dependencies, listing names, locations, and descriptions; and dotnet-package-skills, which scans project packages and copies skills to the .agents/skills directory for inclusion in version control.

Microsoft demonstrated this using the Markout library, which includes five skills covering output formats and content composition. Test results showed varying impacts on AI models of different capabilities: for powerful models already capable of completing most development tasks, adding skills resulted in limited performance improvement; for weaker models, they might fail to consistently follow the skill guidelines. Additionally, poorly designed skill content could increase token consumption, requiring maintainers to evaluate actual benefits.

Currently, NuGet MCP does not yet integrate skill support, but Microsoft plans to add related features in the future. Since AI agents may take actions based on SKILL.md content, Microsoft also warned developers to pay attention to package sources to prevent malicious instructions from becoming a new supply chain attack vector.

Quellen:iThome

Forschung

Linux Software Firm SUSE Cuts 10% of Jobs, CEO Cites AI Automation

Thema · SUSE裁员与AI转型Kurzfassung
2026-11-04 00:14 GMT+8

Linux software provider SUSE has announced a global workforce reduction of approximately 10%, affecting around 190 employees. According to Swiss tech outlet Inside IT, the news stems from an internal memo sent by SUSE CEO Dirk-Peter van Leeuwen to staff, initially reported by UK-based The Register.

In the memo, van Leeuwen stated that automation and AI are fundamentally changing how software companies operate. He wrote, “The capabilities of AI are developing exponentially and forcing us to rethink our work.” Modern tools have enabled teams to work faster while reducing operational overhead and hierarchy levels, leading to a significant decrease in demand for certain existing functions.

It remains unclear which specific roles and locations are affected. SUSE intends to inform impacted employees by November 4. The restructuring is part of a broader strategic review aimed at ensuring long-term sustainability and resilience.

Quellen:inside-it.ch

Forschung

Google's Orbital AI Satellite Transmits First Data, Limited by Power Constraints

Wesentlich
2026-11-04 00:00 GMT+8

According to German tech outlet t3n.de, the prototype satellite from Google's "Suncatcher" project successfully generated and transmitted its first data approximately three weeks after launch. The satellite, equipped with DeepMind's Gemma 4 model, produced a three-line poem (haiku) about space based on ground instructions and sent it back to Earth. This marks the transition of Google's experiment in exploring space as scalable machine learning infrastructure into actual operation.

Despite achieving end-to-end communication, hardware limitations reveal the current bottleneck for orbital computing. The satellite houses four Tensor Processing Units (TPUs), with solar panels delivering up to 1 kilowatt of power. Due to high chip energy consumption, the system cannot drive all TPUs simultaneously, operating only one at a time. Furthermore, to avoid overloading the power supply and cooling systems, computational tasks are executed only in short intervals of about 15 minutes, preventing continuous operation.

Google stated that if the onboard TPUs perform as expected, it plans to launch two more satellites with higher power generation capabilities in 2027 to build a more robust orbital computing network. This progress aligns with visions proposed by companies like SpaceX to place data centers in orbit. Supporters argue that the space environment helps alleviate the electricity and water cooling pressures faced by ground-based data centers, while critics point out the need to consider the carbon footprint associated with manufacturing, launching, and disposal.

Quellen:t3n.de

Forschung

Amazon Ads Announces Launch of 'Amazon Ads Agent' in Japan, Using AI to Automate Channel Selection and Bidding

Thema · 亚马逊广告AI代理Wesentlich
2026-11-04 15:32 GMT+8

Amazon Ads, the advertising arm of U.S. e-commerce giant Amazon, announced on November 4 at its annual "unBoxed Tokyo 2026" conference that it will introduce "Amazon Ads Agent," a new AI-centric advertising platform, to the Japanese market. The platform aims to simplify the ad placement process: advertisers only need to set business goals such as "increase sales" or "expand awareness" and input their budget, and the system will use AI to automatically decide where and at what price to place ads.

The core component of this release is "DVA+" (Digital Video Advertising Plus), which consolidates previously separate purchasing methods for sponsored display ads, sponsored TV ads, and programmatic advertising into a unified system. After advertisers register products and creative assets, the AI automatically distributes ads across streaming TV, online video, display networks, and audio, covering both Amazon's internal sites and external websites, based on shopping, browsing, and viewing signals owned by Amazon. DVA+ also includes "Brand+" for targeting users likely to take next actions and "Performance+" for targeting purchase intent moments. For experienced users who wish to retain control, the platform still offers detailed settings such as frequency caps and manual bidding strategies.

Another highlight is the conversational interface. Advertisers can describe target audiences in natural language, and the system returns category and keyword suggestions while utilizing the data clean room "Amazon Marketing Cloud (AMC)" for analysis. According to Amazon Ads, this feature summarizes results from approximately 1,000 metrics and provides reach predictions, budget allocation suggestions in media planning, and one-click application of recommended bids for sponsored product ads.

Amazon stated that in the U.S. market, advertisers using natural language to set targeting goals reduced their average cost per mille (CPM) by more than 10% while increasing new customers by over 25%. Additionally, the "Geographic Insights & Activation" feature, already available in Japan, automatically adjusts bids based on regional penetration maps; the conference also highlighted a case where Nestlé Purina reduced video view unit costs by 86% in the U.S. using these tools. DVA+ and conversational features are scheduled to roll out sequentially in Japan within the coming months.

Quellen:テクノエッジ TechnoEdge

Forschung

Saudi Aramco Q3 Profit Rises 9% to $31.2 Billion as CEO Sees Multi-Year High Oil Prices

Thema · 沙特阿美Q3业绩Wesentlich
2026-11-03 21:25 GMT+8

Saudi Aramco reported a 9% increase in third-quarter profit, reaching $31.2 billion. Despite lower production volumes due to attacks on facilities by Yemen’s Houthis and Iranian strikes on ships in the Strait of Hormuz, a 35% jump in the average selling price of its oil more than offset the volume decline. The company also increased dividend payments, most of which go to the Saudi state.

CEO Amin Nasser stated on an earnings call that the company aims to maximize production through the end of the year by "utilizing all the routes that we have," including risky ship-to-ship transfers and increased shipments through the Red Sea. He predicted buoyant demand lasting for several years and noted that it could take two years to refill depleted global oil storage reserves even if the Iran war were resolved today.

This financial result confirms that geopolitical risk premiums in oil are translating into substantial cash flow for major energy issuers. However, despite strong earnings, Aramco’s stock has lagged behind global majors that mostly operate outside the Gulf, reflecting persistent investor concerns about regional supply disruption risks.

Quellen:semafor.com

Forschung

Skelter Labs CTO Seo Jong-hoon: The Real Bottleneck for Enterprise AI Agents Is Not Model Performance, But Control

Thema · 스켈터랩스BELLA平台Wesentlich
2026-11-04 15:31 GMT+8

When enterprises attempt to deploy AI agents into actual business operations, the first wall they hit is often not that the models are not smart enough, but the inability to answer a fundamental question: who is using this agent to access what documents and tools, under what conditions? Seo Jong-hoon, CTO of Skelter Labs, summarizes this dilemma as follows: the bottleneck introduced by AI agents is not model performance, but control.

He provides a specific failure scenario: if each team directly calls different large language models, the company level finds it difficult to track where costs occur, and each team's guardrails differ, making it impossible to maintain consistent control levels across the entire organization. This is not a hypothetical issue; it is an actual permission management challenge faced when deploying AI agents after generative AI has diffused within enterprises.

Seo Jong-hoon's argument rests on a judgment: the past AI market focused on training better foundational models, but in the agent era, open models such as OpenAI and Claude have already been validated. Enterprises truly do not need to build their own stronger models; instead, they need to flexibly and securely manage these ready-made models within their own environments.

To ground this argument, Skelter Labs designed the BELLA platform, with the core idea of running the entire lifecycle of AI agents—creation, deployment, and collaboration—on a unified control foundation. The platform consists of four layers: the Identity Layer identifies the currently logged-in end user and passes this information uninterrupted to the final step of agent execution, i.e., search and tool invocation; the Policy Layer uses the user information passed from the Identity Layer to determine access permissions for specific documents, tools, and agents one by one, blocking unauthorized access at the source; the AI Gateway Layer serves as the sole channel for all large model calls and inter-agent calls, centrally executing cost statistics by user and department, budget caps, model routing, and security guardrails; the Asset Layer manages all AI assets including document indexes, MCP tools, workflows, prompts, and skills, dividing sharing and isolation scopes by enterprise, team, and individual levels.

At the defense level, BELLA sets up four checkpoints: Data and Retrieval-Augmented Generation (RAG) integrity protection prevents document chunks from being tampered with and filters prompt injection and hidden characters during registration; Input/Output Guardrails intercept injection attacks and mask Korean-format personal information, API keys, internal IPs, and domains; Agent Behavior Control mandates human approval based on risk levels and executes code in isolated sandboxes to prevent infinite loops or unauthorized tool calls; Gateway Resource Protection limits token quotas and request frequencies by user and key, detecting abnormal consumption patterns.

Seo Jong-hoon further emphasizes that the platform should not be a layer that blocks change, but one that absorbs change. Models and tools are changing rapidly; if the platform binds itself to a single model or cloud provider, it will become a burden during technological iteration. BELLA is built on Kubernetes, supports on-premises deployment and private clouds, and does not rely on specific models or cloud vendors.

Quellen:바이라인네트워크

Forschung

More Than 60% of Eurozone Households Refuse to Buy Stocks Even with a €50,000 Lottery Windfall

Wesentlich
2026-11-04 15:01 GMT+8

A new study based on the Eurozone Consumer Expectation Survey finds that even assuming a €50,000 lottery win, more than 60% of Eurozone households would still not invest in any stocks. Joachim Klement, author of Klement on Investing, interprets this research by noting that a single windfall is far from sufficient to bring Europeans into the stock market.

Behavioral finance research suggests that people tend to invest windfall gains more aggressively than earned income, as these unexpected sums are allocated to different mental accounts. Additionally, individuals with greater wealth typically invest more in stocks because they can absorb larger potential losses. If these two mechanisms hold true in the Eurozone, a €50,000 lottery prize should significantly boost household stock market participation.

The survey data does show evidence of both effects, but to a very limited extent. After hypothetically winning a €50,000 lottery prize, the participation rate among households that did not previously hold stocks increased by only 8.4 to 12.8 percentage points. Meanwhile, the proportion of money invested in stocks by those already in the market hardly changed at all. More critically, even facing such a windfall, over 60% of Eurozone households stated they would not invest in any stocks.

Klement concludes that this clearly demonstrates that simply providing money is insufficient to get Europeans to invest in stocks. This finding implies that the reason Eurozone households do not enter the stock market cannot be simply attributed to a lack of investable capital—because even when given €50,000, over 60% still refuse. However, it is important to note that this study relies on hypothetical survey responses asking respondents what they would do if they won the lottery, rather than observing actual investment behavior following real windfalls. Hypothetical answers may not fully reflect real-world decision-making, but given that the refusal rate exceeds 60%, even with some bias, the directional conclusion remains valuable.

Quellen:klementoninvesting.substack.com

Forschung

South Korean AI Commerce Startup Termit Raises Undisclosed Seed Funding from Mashup Ventures

Kurzfassung
2026-11-04 08:24 GMT+8

South Korean AI commerce technology startup Termit announced it has received seed funding from venture capital firm Mashup Ventures, with the amount undisclosed. The company operates 'Creage,' a comprehensive marketing content agent that uses AI to analyze product features and generate online sales content—including detail pages, thumbnails, banners, and short videos—within three minutes.

Unlike many AI design services that output images, Creage generates detail pages as HTML code with preserved text information. This allows search engines and AI shopping agents to directly read the textual content, optimizing search visibility and adapting to AI-driven shopping environments. The model is reportedly trained on over 50,000 verified commerce content datasets designed to improve purchase conversion rates.

Lee Seung-guk, a partner at Mashup Ventures, stated that Termit acquired more than 15,000 customers and achieved monthly profitability within 10 months of its official service launch, relying solely on partnerships and word-of-mouth. Kim Ji-sung, CEO of Termit, noted that as AI agents begin to participate in finding, comparing, and purchasing products, accurate AI understanding of goods will become a key factor influencing sales, which is why Creage insists on generating machine-readable code.

Quellen:스타트업 생태계의 모든 것 \'플래텀(Platum)\'

Forschung

EU Court Advocate General: AI Chatbots May Infringe Copyright by Retrieving News Articles

Wesentlich
2026-11-04 12:33 GMT+8

According to reports from German tech media heise online and Austrian ORF, Maciej Szpunar, Advocate General at the Court of Justice of the European Union (CJEU), issued a legal opinion on November 3, 2026, stating that AI chatbots like Google's Gemini may infringe on publishers' reproduction and public communication rights when they retrieve and use news articles from websites to provide answers to users.

The opinion was delivered in a case where Like Company, a Hungarian news publisher operating several portals, sued Google. The publisher alleged that Gemini had reproduced and made publicly available its protected texts without authorization. As the case involves interpretation of EU law, the Budapest court referred questions to the CJEU in Luxembourg. Szpunar emphasized that his opinion is not binding on the judges, with the final judgment expected later; previous indications suggested a ruling would likely not come before 2027.

The Advocate General proposed two key criteria for determining infringement. First, it depends on whether the chatbot accessed the article unlawfully, for example by bypassing technical measures such as paywalls. Second, Szpunar noted that press publishers' rights protect not the reported facts themselves, but the "specific expression form" of the publications. This means copyright law should not grant publishers a monopoly on news facts.

The opinion further clarified that the crucial factor is whether the answer reuses this expression form in a recognizable way such that readers no longer have reason to consult the original text. If the expression form is "systematically recognizable" in the answers, the right of public communication is generally considered infringed.

Additionally, Szpunar deemed two questions from the Hungarian court regarding AI model training inadmissible as hypothetical, since there was no evidence that the plaintiff publisher's articles were used to train the Gemini models. However, he noted that under the so-called "text and data mining exception," automated reading and analysis of texts and data for purposes such as recognizing patterns are permissible, providing some legal space for AI training using protected works.

A Google spokesperson responded that the company welcomes the opinion on AI training, stating that "Gemini learns linguistic patterns to generate original content rather than reproducing content," and noting that rights holders have the option to object to the use of their content.

Quellen:heise online

Forschung
Weiter →
Anmelden / Registrieren
LesenForschungRadarAnlageframework

This argument has a clear commercial context: Skelter Labs is selling the BELLA platform, and its partner Metanet Group provides end-to-end services ranging from strategic consulting to infrastructure. The platform architecture description comes from vendor self-reports and has not yet seen independent verification. However, the problem pointed out by Seo Jong-hoon—that multiple teams calling models independently leads to uncontrollable costs and guardrails—is a common realistic obstacle in enterprise AI agent projects.