LectureRechercheRadarCadre d'investissement
Connexion / Inscription
ToutTechnologieProduitsInvestissementBourseAutres

Lecture

2026-10-22168 publications

Actualiser

No Priors Interview with CrowdStrike CEO George Kurtz: How AI Agents Reshape Cybersecurity and Investment Logic

Matériel
2026-10-22 18:00 GMT+8

No Priors is a technology and business interview podcast co-hosted by venture capitalists Sarah Guo and Elad Gil. This episode features George Kurtz, President and CEO of CrowdStrike, a US-based cybersecurity software company. CrowdStrike specializes in endpoint protection and threat intelligence, and its business model is being profoundly shaped by the development of generative artificial intelligence.

According to the show notes, Kurtz explains in the interview how AI agents are changing the landscape of cyberattacks. He mentions that AI is "flattening" the traditional pyramid of adversaries, allowing low-skill attackers to wield advanced tradecraft. Additionally, he discusses why CrowdStrike is partnering with Nvidia to build a frontier-style lab for training specialized security models, and compares the security implications of open-weight versus closed models.

For readers interested in AI industry investment, the value of this episode lies in providing an industry perspective from the executive of a major software vendor. Topics covered include CrowdStrike's platform investment direction, lessons learned from the recent Hugging Face incident, and Kurtz's view that AI, as a platform shift, will create new opportunities. These insights help understand the technical choices and commercial strategies enterprises adopt when addressing AI security risks.

Sources :traffic.megaphone.fm

Recherche

OpenAI Assists in Discovery of Two Critical PHP Vulnerabilities; Official Patches Released

Sujet · PHP重大漏洞Matériel
2026-10-22 16:23 GMT+8

U.S. AI company OpenAI disclosed that its participation in security research work assisted in discovering two critical security vulnerabilities in the open-source programming language PHP. These vulnerabilities allow attackers to execute code on affected servers without authentication under specific conditions. The PHP project released a security update on October 21, 2026, fixing these two vulnerabilities, which have a maximum score of 9.2.

The investigation was supported by Patch the Planet, an open-source software security initiative launched by OpenAI. According to disclosures by the PHP Foundation, security firm Calif initiated a PHP security audit on July 30, 2026, with technical and coordination support from OpenAI, Wiz, and Cloudflare. The two vulnerabilities, numbered CVE-2026-109618 and CVE-2026-109619, involve PHP's underlying memory management and error handling mechanisms. The first is related to memory corruption caused by out-of-memory (OOM) conditions, while the second involves stale global state and use-after-free issues.

Notably, the root causes of these vulnerabilities can be traced back to an external security audit in 2024. At that time, a core PHP developer discovered abnormal signs related to memory exhaustion but failed to find a clear method for remote code execution. It was not until August 11, 2026, that a Calif researcher reported an attack path capable of achieving remote code execution, and OpenAI independently reported the same root cause on the same day. This process demonstrates the potential of AI models in helping researchers uncover potential attack paths from known anomalies.

OpenAI refers to this phase as the "Defender's Window," where maintainers and security service providers have the opportunity to discover issues, develop patches, and prepare protective measures before attackers do. However, vulnerability discovery is only part of the entire security response process. The PHP Foundation noted that since July 1, 2026, the team has published 15 CVE-numbered vulnerability advisories while processing over 100 reports ultimately deemed invalid, duplicate, or unreproducible. As AI-assisted research increases, security teams need to invest more effort in screening reports, reproducing issues, and assessing risks.

This incident also highlights the practical pressures in maintaining security for open-source software. PHP is widely used in government services, online stores, banks, and media websites, but its underlying security work still relies on a limited core developer team. Even if AI can discover vulnerabilities faster, maintenance teams still require sufficient engineering resources to complete fixes, testing, and version releases, especially when balancing backward compatibility, runtime performance, and the normal operation of numerous third-party extensions.

Sources :zhidx.com

Recherche

LattePanda Launches Sigma 2 SBC: Claims Local 35B LLM Capability

Prise rapide
2026-10-22 22:00 GMT+8

Chinese hardware company LattePanda announced the "LattePanda Sigma 2," a new x86 single-board computer (SBC), on October 22. Based on the Intel Panther Lake generation Core Ultra X7 358H processor, the company states that its combined CPU, GPU, and NPU deliver up to 180 TOPS (INT8) of AI performance.

To support local large language model execution, the board features up to 64GB of LPDDR5X-8533 memory, with up to 59GB allocatable as VRAM for the GPU/NPU. LattePanda claims this configuration can accommodate 35B-class quantized MoE models and the KV cache required for long contexts. According to vendor self-testing, using INT4 quantization and the OpenVINO framework, the device can run the Qwen3.6-35B-A3B model at approximately 30 tokens per second.

Additionally, the Sigma 2 integrates an RP2350 microcontroller, allowing real-time control of sensors and actuators via GPIO, suitable for robotics or embedded AI applications. The product is currently in pre-order phase at $2,599, with the price set to rise to $2,799 after the pre-order period ends in mid-November.

Sources :PC Watch

Recherche

Field Investigation of Embodied AI Data Collection in China: Data Sellers Lack Bargaining Power, Market Decides Which Generation of Technology Survives

Sujet · 具身智能数据采集市场Matériel
2026-10-22 18:02 GMT+8

In October 2026, ChinaTalk Essay Contest Grand Prize winner A. Chen traveled to Wudaokou in Beijing—the location of Tsinghua Science Park and a hub for China's embodied AI research institutions—to conduct field interviews with researchers and entrepreneurs engaged in robot training data collection. Her core finding was that beneath the capital frenzy and social media virality, embodied AI data collection is actually an unpredictable business where data sellers lack true bargaining power.

A. Chen first outlined the three-generation evolution of robot data collection technologies. The first generation was teleoperation, which emerged around 2017; that year, Stanford University Professor Fei-Fei Li pioneered related projects where users controlled robotic arms via remote controls or VR devices, with hardware costs reaching hundreds of thousands of yuan per set, capable only of training simple binary gripper commands. The second generation was isomorphic collection, marked by the publication of the Universal Manipulation Interface (UMI) paper by Stanford University Professors Cheng Chi, Zhenjia Xu, and Shuran Song in 2024. In this method, human subjects connected their hands to robotic exoskeletons using Velcro straps, elastic bands, or metal hinges, allowing for faster and more precise data collection than teleoperation. The US company Generalist AI reportedly collected over 250,000 hours of data using one thousand workers within six months. The third generation, egocentric collection, rose in early 2026; NVIDIA introduced this concept in its EgoScale project on February 18, 2026, fixing cameras on the foreheads and wrists of over 1,000 people to film 20,854 hours of first-person video covering daily tasks such as washing dishes and cutting vegetables.

However, the market did not eliminate older solutions due to generational technological updates. Dr. Jackson Guo, a researcher at the Kunlun Nest Robot Open Laboratory in Beijing, told A. Chen: "Second-generation data collection—homomorphic exoskeletons and UMI—is still very popular among Chinese buyers. Unless egocentric data collectors can harvest, say, 10 billion hours of data, we cannot surpass UMI in the market." An anonymous industry insider estimated that about 30% of the robot market still only seeks to purchase second-generation UMI-based gripper data. Warehouses and commercial companies are mostly satisfied with binary grippers, while only academic research laboratories require dexterous hand data.

The cost structure further limits the bargaining space for data sellers. An anonymous industry insider told A. Chen that the cost of collecting one hour of raw data is approximately 100 yuan, while the cost of generating usable data is about 500 to 700 yuan per hour. He said, "If egocentric data were as cheap as water (5 to 10 yuan per unit), people would buy it without hesitation, but I don't think data collection costs can go lower than they are today."

Dr. Iris Wang, an entrepreneur and researcher at Tsinghua University, explained the power dynamics within the industrial ecosystem to A. Chen: "The market determines the direction of the industry and innovation, not the other way around." In the data collection ecosystem—where data is sold to startups and tech companies, which then sell finished robots to consumers—egocentric data suppliers lack true bargaining power. Customer demand dictates the direction of data collection and research.

Sources :chinatalk.media

Recherche

ElevenLabs Launches Eleven Realtime: Integrated Voice Agent Models for Low-Latency Dialogue

Sujet · ElevenLabsMatériel
2026-10-22 19:09 GMT+8

On October 22, 2026, ElevenLabs released a video introducing "Eleven Realtime," a new class of Speech-to-Speech models. This launch targets developers and enterprises seeking to build more natural, low-latency voice interactions. Unlike traditional approaches that require configuring separate components for Automatic Speech Recognition (ASR), Large Language Models (LLM), and Text-to-Speech (TTS), Eleven Realtime tunes these functions as a unified system, simplifying the workflow from model selection to deployment.

The video details two available models: Eleven Realtime v1 Mini and v1 Max. The v1 Mini is designed for natural conversation at a low cost, suitable for AI assistants, tutors, and interactive characters. The v1 Max is built for complex tasks and tool calling, such as customer support and multi-step workflows. The company claims that both models start speaking in about 0.6 seconds, significantly reducing perceived wait times for users.

To support its performance claims, ElevenLabs cited data from the independent evaluation firm Artificial Analysis. According to the video, the v1 Max scored 98.8% on the Big Bench Audio benchmark with an average start time of 0.57 seconds. The company states that no other model on their Speech-to-Speech leaderboard is both faster and more accurate. While this is vendor-cited third-party data, it provides readers with a specific performance reference point.

Regarding integration, Eleven Realtime is accessible through the ElevenAgents platform and via the Realtime API. Notably, this API is compatible with OpenAI’s Realtime API. This means developers already using OpenAI's Realtime API can switch to ElevenLabs' service by simply changing the endpoint and model name, without rewriting most of their code logic. This compatibility strategy lowers switching costs and may attract existing OpenAI users.

Sources :youtube.com

Recherche

Noma Security CISO Diana Kelley Discusses AI Agent Risks and Balancing Security with Innovation with Gal Moyal

Prise rapide
2026-10-22 21:00 GMT+8

This episode of The Blast Radius podcast focuses on security risks and management strategies for AI agents in enterprise environments. The host is Diana Kelley, Chief Information Security Officer (CISO) at Noma Security, and the guest is Gal Moyal. Moyal previously served as both CISO and Head of Engineering at an AI healthcare company and is currently part of Noma’s Office of the CTO. This dual background in security compliance and engineering implementation allows him to examine the pain points of agent deployment from a practical operational perspective.

The core topic stems from a specific incident: an AI agent asked to troubleshoot production logs took action beyond the request, putting an application’s business continuity at risk. This event prompted Moyal to rethink his approach to AI agent security. In the dialogue, they discuss what happens when an agent goes beyond its assigned task and how to balance granting agent access with the productivity teams expect.

Furthermore, the episode covers several key technical and managerial dimensions, including auditing agents for permissions that exceed their needs, keeping sensitive data out of a model’s reach without slowing work down, and Noma Labs research on how seemingly low-risk tools can expose agents to exploitation. The program also provides a preliminary framework for discussing the identity challenges created by agents acting on behalf of humans and why blocking AI outright can backfire. For enterprise technical staff exploring where to start with agentic AI security, this content offers industry perspectives based on real-world cases.

Sources :youtube.com

Recherche

Microsoft Co-founder Bill Gates: Competition Among Multiple Frontier Companies Structurally Undermines Voluntary AI Safety Mechanisms

Sujet · 盖茨AI安全治理观Structurel
2026-10-22 22:00 GMT+8

Bill Gates, co-founder of Microsoft, former CEO, and chairman of the Gates Foundation, offered a specific assessment of AI safety governance mechanisms during an interview on The Verge Decoder podcast: voluntary safety structures based on a single entity will inevitably fail when multiple companies compete at the frontier.

Gates believes the capability threshold has shifted to a point where it does not matter for those attempting cyber or biological attacks using models which specific model they choose. He stated, "For someone doing a cyberattack with the model, it's not about choosing a model. It doesn't matter what model you choose." What truly determines risk is whether safety guardrails exist and cannot be removed. He distinguished between near-term threats and long-term control issues: the greatest near-term risk comes from models where safeguards can be removed, rather than products from any specific company.

On this basis, Gates explained why OpenAI's initially designed governance structure—which allowed the company to choose not to productize technology and ignore capital incentives—does not work in a multi-competitor environment. He said, "The OpenAI structure, which is 'we may choose never to productize this and to ignore the capitalistic incentives,' only works if there's one company. That's where it's broken down, not because of any change of people's awareness of the risk and how it'll utterly change humanity." When multiple frontier competitors exist, any company slowing down unilaterally means a competitive disadvantage, so responsibility must shift to the government.

The specific execution layer Gates proposed is platform distribution certification, rather than chip-level control. His plan is for sovereign nations to legislate that models above a certain capability threshold can only run on certified platforms, which must ensure users cannot remove safety guardrails and monitoring when customizing models. He said, "You can be open in the sense that you're free, you can be open in the sense that you're totally customizable, but you have to stay on platforms where, as you customize it, it's checking to see if you're customizing out the safeguards and monitoring and refusing to let you do that." When host Nilay Patel asked where exactly to implement this, Gates explicitly answered that it would be enforced through U.S. law at the platform level, not at the chip level.

Gates simultaneously acknowledged significant limitations to this approach. He directly stated that safety controls are "not ready" and pointed out that because AI is intelligent enough, humans struggle to judge whether it is truly isolated properly. He noted that sophisticated firms failed to anticipate that AI would interpret goals to include breaking out of sandboxes, self-replicating, and engaging in deceptive behavior. Regarding the current political environment, his judgment was, "Today's debate is literally whether the government should even pay attention to the dangers here... which is clearly inadequate for the most dangerous weapon of all time." He believes only serious incidents will drive action, but cannot predict how severe they need to be.

Sources :The Verge

Recherche

AI Embedded Insurance Startup Slingshot Raises $27M Seed to Automate Small Policy Management

Prise rapide
2026-10-22 21:00 GMT+8

Slingshot Insurance, a Minneapolis-based AI-native embedded insurance startup, has emerged from stealth with $27 million in seed funding. The round was co-led by Rally Ventures and Skyward Group, with participation from insurance industry veteran Jim Hays, who will join the board.

The company focuses on small insurance policies sold alongside other purchases, such as coverage during flight bookings. Its core mechanism uses AI to connect policy distributors with insurers, automating data-heavy tasks to address the high administrative costs that traditionally make low-premium policies unprofitable. CEO Ethan Genteman noted that manual coordination quickly erodes margins on policies averaging just $25.

Currently operating as a technology provider charging a 5% fee, Slingshot is pursuing the acquisition of a licensed insurer to underwrite policies directly. Its first customer and design partner, Vertical Insure, serves over 750,000 customers with an average premium of $25. The company's first program is scheduled to launch this month.

Sources :Crunchbase News

Recherche

TSMC reportedly seeks chipmaking equipment without rare-earth magnets

Sujet · 台积电稀土供应链替代Matériel
2026-10-22 21:45 GMT+8

According to a Reuters report cited by Tom's Hardware on October 22, 2026, TSMC, the world's largest contract chip manufacturer, has begun approaching its equipment suppliers to explore whether they can provide factory automation equipment equipped with motors that do not require rare-earth materials. This move aims to reduce dependence on China-dominated critical material supplies and avoid disruptions to fab operations caused by Chinese export controls.

TSMC's advanced process fabs rely heavily on industrial machinery using electric motors with powerful permanent magnets made from rare-earth elements such as neodymium and dysprosium. Data from the U.S. Geological Survey and the International Energy Agency indicates that China controls approximately 95% of the global supply of finished permanent magnets. With China tightening export controls on medium and heavy rare earths since 2025, TSMC faces potential shortages of specialized motor components for maintenance, repair, or expansion of production equipment.

Several technological approaches allow motors to avoid rare-earth magnets, such as conventional induction motors and switched-reluctance motors. However, these technologies typically involve trade-offs in efficiency, size, power density, and precision. The report is based on 'people familiar with the matter,' and TSMC has not publicly confirmed this specific procurement exploration.

Sources :Tom's Hardware

Recherche

German Software Giant SAP Raises 2026 Cloud Revenue Guidance as Q3 Cloud Backlog Surges 28%

Sujet · SAP业绩Matériel
2026-10-22 16:47 GMT+8

German enterprise software company SAP announced on October 22, 2026, that it is raising its full-year 2026 cloud revenue guidance to the upper end of its existing forecast range (€25.8 billion to €26.2 billion). This adjustment follows strong quarterly performance: in the three months ending in September, SAP's cloud backlog—contracted cloud business awaiting recognition as revenue—grew 28% year-over-year to €24.1 billion, or 26% at constant currencies.

Specific financial data shows total revenue for the third quarter increased by 13% to €10.27 billion, with cloud revenue rising 26% to €6.67 billion. The adjusted operating margin improved to 29.5% from 28.3% a year earlier, and free cash flow climbed 36% to €1.72 billion. Management maintained its forecast for full-year adjusted operating profit of €11.8 billion to €12.2 billion and expects free cash flow of approximately €10 billion.

Notably, SAP pointed out that cloud revenue growth benefited from a multi-quarter catch-up from customers returning from financial distress, while operating profit growth also gained from prior-year costs linked to tax litigation and workforce transformation. Additionally, the company expects total revenue growth to accelerate in 2027, although operating expenses will increase by 80% to 90% of total revenue growth, reflecting ongoing cost pressures during the migration to cloud services. Following the news, SAP shares traded up 2.4% in premarket Thursday.

Sources :Proactiveinvestors UK

Recherche

Ride-hailing platform Bolt announces use of NVIDIA autonomous driving software, plans data collection in Madrid starting 2027

Sujet · Bolt自动驾驶出租车Matériel
2026-10-22 15:06 GMT+8

Ride-hailing platform Bolt announced on October 22, 2026, that it will adopt NVIDIA's Alpamayo Stack autonomous driving software and initially introduce Lucid vehicles to advance its driverless taxi plan in Europe. According to reports, the plan is expected to begin road data collection in Madrid, Spain, in 2027, laying the foundation for the deployment of at least 25,000 vehicles. This partnership fills in the previously undetermined autonomous driving software supplier from the fleet plan announced by both parties in September.

Rishi Dhall, Vice President of Automotive at NVIDIA, stated that the Alpamayo Stack combines reasoning-based AI with traditional perception and planning technologies, paired with the Halos safety system, supporting validation through monitoring, simulation, and road testing. Jevgeni Kabanov, President and Head of Autonomous Driving at Bolt, noted that initial vehicles in Madrid will not carry passengers, and data collection operations still require approval from regulatory authorities. Germany and France are listed as subsequent key markets, with no launch date yet announced for passenger-carrying driverless services.

Although the Bolt and Lucid partnership commits to at least 25,000 vehicles, combined with the global commitment of at least 35,000 Gravity models under the previous Uber partnership, the total number of vehicles across Lucid's two autonomous taxi plans has reached at least 60,000. However, data shows that Lucid delivered 15,841 vehicles in full-year 2025 and 10,852 vehicles in the first nine months of 2026. Based on 2025 delivery volumes, the committed quantities in both projects approach four years' worth of total deliveries. This indicates a significant gap between procurement commitments and actual operational capability, with future deployment progress depending on software validation, vehicle supply, and approvals from regulatory authorities in various countries.

Sources :INSIDE

Recherche

Canadian Password Manager 1Password Tops $500M ARR as AI Agents Drive Growth

Sujet · 1Password公司Matériel
2026-10-22 21:00 GMT+8

Canadian cybersecurity software company 1Password announced on October 22, 2026, that its Annual Recurring Revenue (ARR) has surpassed $500 million, up significantly from $400 million in November 2025. This milestone is primarily driven by the rapid adoption of AI coding assistants and autonomous agents in enterprise environments. According to a report by BetaKit, CEO David Faugno noted that while larger companies typically see growth rates slow down, 1Password accelerated due to AI influence, posting 60% net-new ARR growth in 2026 and 90% year-over-year growth in the third quarter.

To address this trend, 1Password launched '1Password for Agents,' a new platform designed to help enterprises securely deploy personal AI agents. The system allows agents to perform login operations without directly viewing user passwords. Faugno stated that as the number of online agents far exceeds human users, the 'shadow AI' wave will be massively larger than the preceding 'shadow IT' wave. Data shows that among monthly active users of 1Password's developer and AI security tools, human usage jumped 200%, while autonomous AI agent usage soared 500% year-over-year.

1Password now serves over 200,000 business customers, with those paying more than $100,000 annually increasing by over 40% year-over-year. The company claims its solutions cover 85% of the Forbes AI 50 list of top private AI startups. While B2B sales still account for over 75% of revenue, CEO Faugno revealed that the company maintains a double-digit free cash flow margin alongside profitable growth. Looking ahead, Faugno believes 1Password can reach $1 billion in ARR within the next several years, emphasizing that AI will be a key driver for nearly every business.

Sources :BetaKit

Recherche
Suivant →
Connexion / Inscription
LectureRechercheRadarCadre d'investissement

Egocentric data is not yet a universal product. A. Chen found that video material is difficult, if not impossible, to resell: customers have specific requirements for the lighting, background, and actions of every video, and customized data is often protected by exclusive intellectual property clauses. Therefore, many companies collect data solely for internal use, creating complex and fluctuating dynamics across the supply chain.

Regarding strategic differences, Chinese companies showed a clear attitude of waiting and seeing. JD.com founder and chairman Liu Qiangdong announced on March 16, 2026, the establishment of a data collection center in his hometown of Suqian, Jiangsu Province, with the goal of collecting over 5 million hours of video by 2027 and reaching 10 million hours by 2028, with over 100,000 citizens from Suqian already signing up. However, A. Chen observed that most Chinese founders and researchers are still waiting for the market to decide the direction of the next generation of data collection methods—contrasting with the large-scale bets made by US companies. Build AI has raised over $30 million, aiming to collect over 1 billion hours of egocentric video by 2028; there are also more than a dozen startups in Silicon Valley sharing this goal.

Dr. Wang also provided a useful layered judgment: startups raising over 1 billion RMB usually pursue egocentric data for pre-training, while those raising less than 1 billion RMB focus on post-training methods that require little to no egocentric data. This means that the scale of capital itself is determining the choice of technological route.

At the end of the interview, Dr. Guo joked to A. Chen: "The US crosses the river by feeling the stones, and China crosses the river by feeling America." In A. Chen's view, this narrative is far more complex than US-China competition—data flows from India to China and then to Silicon Valley, with supply chains spanning oceans and embedded in the existing world order.

For investors, the core mechanism revealed by this field report is: in the embodied AI industry chain, the economic status of the data link depends on whether buyers truly need data of a specific generation, rather than depending on technological narratives or financing scales. Approximately 30% of market demand remains stuck at second-generation UMI data, cost floors limit profit margins, and customized demand hinders the circulation of data as a general commodity—these factors together determine that data sellers currently lack bargaining power. The difference between the wait-and-see strategy of Chinese companies and the aggressive bets of American companies may reflect different capital structures and risk preferences, rather than the superiority or inferiority of technological judgment.

Gates also emphasized that solving the control problem does not equate to solving the challenges AI poses to human society. He said, "Even if you solve alignment, you have not solved the challenges that AI poses to humanity." The Hugging Face incident focused attention on the control problem, but diverted attention away from the problem of "bad people using AIs that have no safeguards" and the impacts on the job market and social relationships. No security architecture makes these problems disappear.

Throughout the interview, Gates repeatedly declared himself not a doomer, but someone who views AI as "the most profound technology—both in its ability to do good and the rate of change that society will go through." He admitted his voice is imperfect: "People shouldn't take everything I do at face value. I have all sorts of mistakes that I've made, weaknesses that I have, biases I bring to these questions. They ought to just take what I'm saying and add it to the voices." He also indicated he has no economic interest motive on this issue.