DAILY BRIEFING The CAIO desk · Tradesolution SATURDAY 21 NOVEMBER 2026
AI News stream
The frontier, the noise, and everything worth Pedro's five minutes
Story of the day · Frontier labs

Tencent releases Hy4 model weights with million-token context

Tencent releases Hy4 model weights with million-token context
Tencent has released the weights for its flagship Hy4 model, allowing developers to download and deploy it themselves. The model is now available under the Apache 2.0 license, which permits commercial use and modification. Hy4 supports understanding of text, images, and video.
Licences Evaluate whether Hy4 can replace existing API solutions to reduce costs and increase control, as the Apache 2.0 license allows free commercial use.
Image: TechNode
Stories kept
99 / 799 assessed
Filtered out as not AI-related
700
Active sources
99 across 12 categories
Stories by category
Top stories today
Vulnerabilities · 9
Researcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced.
A researcher tested the AI supply chain by planting unique, trackable secrets to demonstrate that data does not remain private. This has revealed a significant privacy disaster. While active exploitation is not confirmed, users should follow the vendor's security advisory to address the issue.
Developer tools · 20
More unveils TIDE framework to boost LLM inference efficiency (translated from ko)
More has unveiled TIDE, a framework designed to enhance the inference performance and efficiency of large language models. The technology was first presented at the Supercomputing 2026 conference and uses data from actual inference runs to continuously improve an auxiliary model. This allows the AI inference system to autonomously identify the optimal execution method in response to changing user requests.
Local models · 2
Welcome Inkling 2 Mini, Cutting-edge Performance for your Local Setup
Inkling 2 Mini is a new model designed for local execution on user hardware. It focuses on delivering high performance in a compact size, making it suitable for local setups.

Pick a category to read the stories, or show all categories.

Vulnerabilities9
Researcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced.
A researcher tested the AI supply chain by planting unique, trackable secrets to demonstrate that data does not remain private. This has revealed a significant privacy disaster. While active exploitation is not confirmed, users should follow the vendor's security advisory to address the issue.
Governance Review and update AI vendor risk management to ensure privacy requirements and security guidelines are enforced in contracts.
EU-UK Summit: AI-assisted cyberattack on European Commission (translated from de)
The European Commission has taken its personnel platform offline following a reported AI-assisted cyberattack. It is unclear whether data has been exfiltrated. Readers should follow the European Commission's security advisory for further updates.
Governance Review and update internal cybersecurity and AI safety guidelines to prepare teams for the increased risk of AI-driven attacks on critical systems.
Also today · 7
The vulnerability affects Azure API Management (APIM). It is not confirmed that the vulnerability is actively exploited. NSM has shared measures to address the threat.
A spy court has criticized the UK government's secrecy and forced it to admit to receiving an Apple ‘backdoor’ demand. It is not confirmed that the vulnerability is actively exploited. Readers should follow Apple and UK authorities' security advisories for further details.
An optional update fixes a misleading Android error for versions 25H2 and 26H2. It is not confirmed that the vulnerability is actively exploited. Users should install the optional update to resolve the issue.
The hacker group The Gentlemen claims to have stolen confidential data from Nvidia. Exploitation is not confirmed, as the incident involves a ransom demand rather than active use of the data. Nvidia should follow the vendor's security advisory and guidelines to address the threat.
Critical Virtzilla bugs allow authentication bypass, as a nested virtualization issue got Digital Ocean ducking. Exploitation is not confirmed in the source text. Users should follow the vendor's security advisory for updates.
Broadcom VMware vCenter is affected by an authentication bypass vulnerability in the VMware Directory Service. The vulnerability is actively exploited, as it is listed in the CISA KEV. Users should follow Broadcom's security advisory to apply updates.
Cisco Expressway Series is affected by a vulnerability that allows unauthenticated remote attackers to execute commands as root. The vulnerability is actively exploited and is listed in CISA KEV. Users should follow the Cisco security advisory to apply updates.
Developer tools20
More unveils TIDE framework to boost LLM inference efficiency (translated from ko)
More has unveiled TIDE, a framework designed to enhance the inference performance and efficiency of large language models. The technology was first presented at the Supercomputing 2026 conference and uses data from actual inference runs to continuously improve an auxiliary model. This allows the AI inference system to autonomously identify the optimal execution method in response to changing user requests.
OpenAI rolls out dots agent to select ChatGPT Plus users (translated from ja)
OpenAI has started offering its always-on agent, dots, to a subset of ChatGPT Plus users. The feature is also being made available to all users of ChatGPT Business Standard.
Licences Assess whether existing ChatGPT Business Standard licenses automatically include access to the dots agent, and document this to ensure we fully leverage the value of our licenses.
Check what your AI agents do before they do it, with Semantic governance policies
Semantic governance policies enable developers to govern AI agents by evaluating every proposed tool call inline before execution. Operating via Agent Gateway, the policy engine checks each action against user intent and plain-language rules to prevent unauthorized data sharing and prompt injection attacks. Developers can easily attach natural-language constraints to their agents and monitor auditable verdicts through Cloud Logging.
Governance Evaluate implementing semantic governance policies to ensure AI agent actions comply with internal rules and risk assessments.
Also today · 17
The Claude Code update introduces a new built-in module, easel, that draws charts based on MCP metrics. It also enhances enterprise sign-in by supporting organizations managed in the Anthropic Enterprise Admin Console. Additionally, the release adds several new metrics and configuration options for self-hosted runners and the VSCode integration.
Google is bringing "Computer Use" to Gemini for Mac, allowing the AI to operate desktop apps, local files, and websites in an isolated background window with a live preview. This feature enables users to continue working in other applications while the AI completes tasks.
Also covered by: Computer Sweden, AI타임스
Meta has developed Triton Low-Level Extensions (TLX) to bridge the gap between high-level GPU programming abstractions and low-level execution control. The extension aims to fully exploit modern GPU architectures by providing access to advanced hardware capabilities.
Google has created "Mirror Dimension," an isolated copy of the internet where AI agents can practice web surfing without causing real-world issues. The environment features realistic websites such as social media and banking services, as well as simulated cyberattacks like prompt injections. Google plans to let more internal teams use the tool and hopes it will inspire other developers to build similar test environments.
Google is using an isolated test environment to expose its new Chrome Browseragent to manipulated content. The goal is to assess the agent's security by having it encounter thousands of fake websites.
Google has announced it is donating Agent Substrate and gVisor to the Cloud Native Computing Foundation (CNCF). This places two key components of its AI agent infrastructure under vendor-neutral open-source governance.
Companies will soon be able to run the open-source Hermes Agent from Nous Research directly inside Microsoft Teams. The Collective Wisdom feature allows one employee's AI prompts and workflows to be shared with the entire team.
Alex Spinelli, Senior Vice President of AI and Developer Platforms at Arm, discusses bringing ML capabilities to edge devices. He explains how software tools must adapt to the demands of autonomous agents and why functional safety is critical as AI enters the physical world of robotics.
Meta has announced that all users, including those on free plans, can now create a dedicated email address for the AI agent Muse. The address acts as a proxy, allowing Muse to handle incoming and outgoing emails on the user's behalf.
OpenAI has revealed an internal system called Performance Factory that leverages Codex to automatically identify and resolve performance issues in ChatGPT. The system analyzes logs and code to pinpoint the causes of slowdowns, proposes fixes, and verifies that the issues are resolved after deployment.
AWS has released AWS Emulator, a free, CC0-licensed local stand-in for more than 200 of its services, generated from its public API models and aimed at the test loops of coding agents. AWS says agents wrote and maintain all of the code and publishes per-release fidelity scores. The docs say it does not evaluate IAM, and practitioners are split on trusting an emulator nobody reviewed.
Microsoft has announced the launch of Microsoft Perception, a new service that uses AI agents to counter cyberattacks driven by artificial intelligence. The service was unveiled at the annual Microsoft Ignite 2026 event.
Harness has introduced a new feature that allows developers to adapt their preferred language models to the platform's harness. The goal is to improve both performance and cost efficiency by optimizing the interaction between the model and the infrastructure.
Agent Anomaly Detection now monitors ADK agents running on AWS, Azure, or on-premises by analyzing their traces and logs within a Google Cloud project. By ingesting OpenTelemetry-formatted data through the Telemetry API, it evaluates full conversations for behavioral drift without requiring access to the agent's underlying runtime. This allows security teams to detect risks such as rogue behaviors, tool misuse, and resource exhaustion across multi-cloud environments.
The article explains how to privately connect Gemini Enterprise to internal StreamableHTTP MCP servers running on Cloud Run. The solution utilizes Egress Agent Gateway, Private Service Connect, FQDN-scoped Private Cloud DNS, and dual-header OAuth 2.0 authentication to ensure secure connectivity.
When building a custom enterprise UI on top of Gemini Enterprise's :streamAssist API, developers must manually implement three core authentication layers. These include authenticating users or workloads via Google Identity, managing third-party 3-legged OAuth for Federated Connectors through the Keystore, and configuring on-demand per-tool OAuth for custom ADK agents using Vertex AI Agent Engine and GCP Auth Manager.
The article explains how to enable and configure the computer use feature for organizations using Team and Enterprise plans. It is aimed at developers and administrators looking to integrate this capability into their projects.
Local models2
Welcome Inkling 2 Mini, Cutting-edge Performance for your Local Setup
Inkling 2 Mini is a new model designed for local execution on user hardware. It focuses on delivering high performance in a compact size, making it suitable for local setups.
Licences Evaluate if Inkling 2 Mini can replace existing cloud licenses to reduce costs and improve data security.
Also today · 1
The new version of Ollama adds support for Thinking Machines' Inkling 2 Mini model, which runs efficiently on a Mac with about 20 GB of memory. The update also improves concurrent request handling on Apple Silicon and fixes several bugs related to GPU memory and API handling.
Frontier labs7
Google DeepMind outlines four development paths beyond AGI (translated from ko)
Google DeepMind has presented four technical pathways for the continued evolution of artificial intelligence after the achievement of AGI. Researchers at the lab argue that even once AI reaches human-level intelligence, development will not stop, but will progress toward systems that surpass the capabilities of groups of human experts.
Tencent releases Hy4 model weights with million-token context
Tencent has released the weights for its flagship Hy4 model, allowing developers to download and deploy it themselves. The model is now available under the Apache 2.0 license, which permits commercial use and modification. Hy4 supports understanding of text, images, and video.
Licences Evaluate whether Hy4 can replace existing API solutions to reduce costs and increase control, as the Apache 2.0 license allows free commercial use.
Meta unveils Self-Improving Memory for AI agents (translated from ko)
Meta has introduced a new technology called Self-Improving Memory that allows AI agents to autonomously refine how they manage and retrieve information. The system adjusts its own guidelines based on experiences from previous tasks, which can also be used to enhance the model's learning. This follows recent reports from OpenAI and Anthropic regarding advancements in self-improving AI systems.
Also today · 4
OpenAI has significantly strengthened safeguards against extracting internal reasoning from GPT-6.2 Astra and GPT-6.2 Sol. In adaptive red-teaming tests, the success rate for extracting protected reasoning fell from 54 percent to 0.2 percent. The company is also suspending thousands to tens of thousands of accounts each week over coordinated distillation campaigns.
Meta has confirmed that its new AI agent, Muse, mixed up conversations between different users. The company attributes the errors to hallucinations and states that these are isolated incidents.
OpenAI has introduced a layered approach to combat adversarial distillation. The strategy combines model training, safeguards, detection, and enforcement to protect against capability extraction.
Google DeepMind has chosen to name its new Gemini models after chemical elements, sparking significant interest among users. The company has also introduced a feature allowing users to select models based on the periodic table, while leadership has confirmed that further releases are on the way.
Industry news21
Mondelez Scales Enterprise Generative BI Tool Under CIDO Catalano
Mondelez is expanding the use of its enterprise generative business intelligence tool. Under the leadership of CIDO Catalano, the company has established a more structured technology roadmap focused on a common foundation and specific vectors of change.
Licences Evaluate whether Mondelez' strategy of using a shared platform for generative BI could give us access to more favorable licensing terms or shared costs for tools we already use.
Lessons in AI ROI from Coty, Clorox, and Other CPGs
The article examines what separates AI experimentation from meaningful business results. It draws on experiences from major consumer packaged goods companies like Coty and Clorox to identify factors that drive effective return on investment in artificial intelligence.
Measurement Analyze the article to identify the specific KPIs and metrics used by Coty and Clorox to measure AI ROI, and consider implementing similar metrics in our own reporting.
The most misunderstood graph in AI just got a major update
A new report from the AI evaluation nonprofit METR finds that some AI capabilities are continuing to improve along an exponential trajectory. The report updates the organization's widely discussed time horizon plot, while also indicating that models are not yet as effective at tasks requiring judgment as they are at coding.
Measurement Incorporate METR's findings on exponential growth and judgment limitations into the next quarterly assessment of AI adoption and risk profile.
Also today · 18
Meta is re-evaluating its 2030 sustainability goals due to record-high emissions associated with AI data center operations. The company has begun to reassess its commitments to achieving net-zero emissions by 2030.
A new report from Marsh finds that executives receive significantly more AI training than other employees. Researchers suggest this disparity may explain why many employers are experiencing a stalled transformation in their adoption of artificial intelligence.
Anthropic has been valued at 12 trillion kroner following the largest IPO in history. With a massive deficit and new shareholders, the question is whether the company's ideals will survive.
Following the launch of AI agents by Meta, OpenAI, Anthropic, and Google, Apple is now developing its own solution. At the same time, the company is introducing new restrictions that make it harder for rivals to access its devices.
Nvidia is reportedly close to acquiring AI startup Inception for at least $500 million. The two-year-old company develops language models based on diffusion processes, an alternative to the token-by-token generation used by models like ChatGPT. The deal is expected to bolster Nvidia's open-source Nemotron models and its push into robotics.
Also covered by: AI타임스
Google is rolling out a collapsible sidebar globally on its Desktop Search homepage. In AI Mode, it lets users reopen recent AI searches, search threads, and create notebooks, a feature designed to make it easier to continue longer web research.
Also covered by: Ars Technica AI
Project Suncatcher is entering its next phase, with plans to launch several satellites equipped with Google's AI chips into orbit by 2029. Blue Origin will handle the launch of these satellites, which are intended to serve as data centers in space.
Microsoft has introduced Copilot Workspaces, a new collaboration environment where employees and AI agents can work together on shared projects. The tool replaces Copilot Notebooks and adds live indicators for colleagues as well as support for Autopilot agents that run in the background. The goal is to position Copilot as a central hub for team productivity by sharing context between humans and agents.
Investors demand SpaceX keep Nvidia GPUs on EarthGOLEM.DE - KUENSTLICHE INTELLIGENZ
SpaceX plans to use new funding to purchase Nvidia GPUs. However, investors require that these units remain on Earth as collateral rather than being sent into space.
OpenAI has introduced a new pricing structure that allows developers to save up to 40 percent on API costs by accepting longer response times. Setting a completion target of two hours reduces the price by 30 percent, while an eight-hour target offers a 40 percent discount. This provides companies and developers with an opportunity to lower costs for large-scale AI tasks.
Anthropic, the company behind the AI service Claude, debuted on the stock market on Thursday. It is the largest IPO in history.
Moonshot AI has cancelled a $150 million contract to lease Nvidia Blackwell Ultra B300 chips through an Indonesian data center. The decision follows new US regulations that restrict Chinese companies from accessing advanced AI hardware via overseas data centers.
Google has signed a contract with Blue Origin to test a solar-powered AI data center in orbit. The company plans to launch several satellites equipped with its own AI chips in the first quarter of 2029 to evaluate the feasibility of large-scale AI inference in space.
Meta is planning to launch the wearable AI device Muse Charm for $300 in December. The device features the Muse AI agent and includes up to 12 months of free 5G data from Verizon, allowing users to access the AI assistant without a smartphone or Wi-Fi connection.
A survey commissioned by Avanade reveals that 77 percent of business decision-makers believe their organizations are focusing too heavily on incremental improvements to existing processes. Only 11 percent expect AI to fundamentally transform how their organizations operate, create value, or compete in the coming year. According to Avanade, this is often because AI is being bolted onto outdated processes rather than used to reshape work methods.
Claire Vo, founder of the AI platform ChatPRD, reports widespread fraud where groups abuse free trials to resell tokens or run commercial projects. The scammers use specific prompts to identify underlying models, prompting ChatPRD to discontinue its free trials entirely.
Hebbia tested GPT-6 Astra for financial research. Its prerelease evaluations found gains in source coverage, citation relevance, and presentation quality.
MedChemBench evaluates how well AI models interpret molecular structures and reason from experimental evidence. The benchmark also tests their ability to select compounds in early drug discovery.
Europe6
OVH launches its own AI model and office suite (translated from fr)
French data center operator OVH is expanding its business by launching its own artificial intelligence model and an office suite. The strategy shares similarities with the approach of the start-up Mistral AI.
Licences Evaluate whether OVH's new AI model and office suite can replace existing vendors to reduce costs or increase data control.
UK AI minister rejects calls for sovereign model, sparking backlash: 'I'm so angry'
The UK AI minister has rejected proposals to develop a national, sovereign language model. The statement has triggered strong reactions within the industry, with critics expressing anger over the British approach to AI sovereignty.
Governance Assess how the UK's rejection of a national sovereign model impacts our strategic choice between commercial APIs and self-hosted solutions, particularly regarding data residency and vendor dependency.
Also today · 4
Many content creators are leaving Etsy after the marketplace changed its terms of service to allow the use of their art and content for training AI models. This backlash follows shortly after Etsy decided to crack down on AI-generated content on its platform.
Ubisoft has established a subsidiary called Motherbrain to develop AI solutions for gameplay mechanics. The company plans to offer these tools to other game developers while simultaneously laying off 166 employees at its Canadian studios.
Data from INE shows that AI usage among Portuguese people aged 16 to 74 has increased by 15% compared to last year. Young people aged 16 to 24 lead in adoption, primarily for personal purposes.
Deutsche Telekom is taking a stake in the French AI company Mistral AI for 100 million euros. The investment aims to expand both companies' AI business in the German market.
Nordics4
RAM memory prices surge due to AI demand (translated from sv)
Demand for RAM memory has surged as artificial intelligence infrastructure expands, leading to skyrocketing prices. Professor Kenneth Carling at Högskolan Dalarna explains how this situation developed and when prices are expected to drop again.
Licences Consider negotiating longer lock-in periods or fixed-price contracts for IT hardware and cloud services to mitigate the impact of rising RAM prices.
Also today · 3
Eidskog municipality is implementing artificial intelligence to automate the writing of medical records during patient meetings. The initiative aims to save time for healthcare staff amid tight budgets and a growing elderly population.
Professor Haakon Bryhni sold the Tydal data center to a Chinese crypto billionaire but will continue to lead the project to convert it into Europe's largest AI factory.
Professor Haakon Bryhni sold the Tydal data center to a Chinese crypto billionaire but will lead the project to convert it into Europe's largest AI factory. The facility is set to cost approximately 50 billion NOK to build.
China and Asia18
Unknown AI model 'Gestalt' outperforms human designers in blind test (translated from ko)
An AI model named 'Gestalt' has taken first place in a design evaluation that compared it against human experts and models from major AI companies. In a blind test conducted by Contra Labs and BenchFlow, 'Gestalt' received higher scores than the designs created by professional designers.
Measurement Evaluate whether 'Gestalt' can replace or supplement existing design tools by running an internal pilot with specific design tasks to measure quality gains.
Neurofusion ties for first place on AI forecasting benchmark (translated from ko)
Korean AI company Neurofusion has achieved a joint first-place ranking on the global ForecastBench benchmark. The company outperforms major players such as Google DeepMind, OpenAI, and Anthropic in predicting economic, financial, and international events.
Measurement Evaluate whether Neurofusion models can replace or supplement existing prediction tools to improve the accuracy of our financial forecasts.
GPT-6.2 Astra and Gemini 4 Neon score perfect on Suneung, LG AI Research's K-Exaone 2.5 tops Korean models (translated from ko)
OpenAI's GPT-6.2 Astra and Google's Gemini 4 Neon achieved perfect scores across all subjects in the 2027 Korean College Scholastic Ability Test (Suneung). Among domestic models, LG AI Research's K-Exaone 2.5 recorded the highest performance. The results were published by a student at Chung-Ang University who developed a dedicated LLM benchmark for the exam.
Measurement Assess whether the Suneung benchmark is relevant to our specific use cases and use the results as a reference point when comparing the performance of different LLM providers.
Also today · 15
Major property insurers in Japan are reporting a sharp jump in cyber insurance claims. This surge is driven by a nationwide increase in hacking incidents powered by artificial intelligence.
Brewer Asahi has announced an $85 million investment in artificial intelligence to strengthen its cybersecurity. The upgrade is a direct response to a recent ransomware attack on the company.
Japan's sovereign AI initiative is facing a $5 billion cost increase. This surge is driven by rising chip prices and a weak Japanese yen.
A Bloomberg News analysis of 10 million job postings suggests AI is compounding China's already weak labor market. Entry-level workers and the creative industry are feeling the shift most, with more than 90% of Chinese micro-dramas now produced using AI. Former central bank adviser Cai Fang warns that a larger AI shock is still ahead.
Pragmatik Labs, the AI startup founded by former Alibaba Qwen leader Lin Junyang, has raised new funding at a $5 billion valuation, Bloomberg reported on Friday, citing people familiar with the matter. The round raised tens of millions of dollars, with new investors including HyT Capital, CDH Investments and automaker Geely.
Alibaba is using chips developed by its semiconductor subsidiary T-Head to train its next Qwen model, CEO Eddie Wu said during the company’s quarterly earnings call. The disclosure extends T-Head’s role across both major stages of AI computing: training models and providing inference services on the Bailian platform.
The student counseling department at Kyoto University has published a document titled 'Things to be careful about when consulting AI about your problems.' The checklist advises students to take a break from using AI if they recognize certain signs of dependence.
Ant Group has expanded the open ecosystem of its AI assistant, Lingguang, by adding seven new external services, including WeCom and Tencent Docs. This brings the total number of supported external services to 54, enhancing the assistant's ability to handle tasks in both professional and personal contexts. Since its launch, Lingguang 2.0 has executed over 11 million tasks, highlighting the company's focus on shifting AI from simple dialogue to active task management.
A best paper winner at ACM MM 2026, featuring a second-year undergraduate from Xidian University as the first author, has faced serious allegations of academic misconduct. Critics have pointed to incorrect references, suspicious data, and weak comparisons, sparking a debate about the integrity of the work. The corresponding author has responded to the accusations, but the case has generated significant attention in the academic community.
The Korean Ministry of Science and ICT has published draft regulations defining the certification criteria for AI Data Centers (AIDC) under the new AIDC Special Act. To qualify for government benefits, data centers must meet specific thresholds for AI accelerator scale, power capacity, and cooling infrastructure. The regulations are open for public comment until December 30, ahead of the law's implementation in March 2026.
Hitachi will launch a new digital entity, Hitachi Digital, in April 2027 by merging two US subsidiaries. The company, which will employ approximately 35,000 people across 27 countries, aims to double its business size within five years. Its strategy involves tripling its field development engineer (FDE) team to 3,000 within a year and focusing on its top 200 to 300 customers to secure dominance in physical AI.
Researchers at China Huaneng Group have developed an AI model to detect early-stage fire and smoke in solar power plants. The model, named PVFire-YOLO, is an improved version of YOLOv12 designed to minimize false positives caused by reflected sunlight and missed faint smoke. Testing showed a 62.9% reduction in false alarms per camera per hour compared to existing models.
Qi Lu, a former senior research scientist at ByteDance, has been appointed head of the Insta360 Research Institute and concurrently serves as the company's Chief AI Officer. He will oversee algorithm and AI development, focusing on integrating artificial intelligence into Insta360's imaging products.
SenseTime chief scientist Lin Dahua explains that the company deliberately invests 20% of its resources in pre-research without immediate product output to secure long-term competitiveness. He argues that the industry is moving away from blind parameter scaling towards more efficient data usage and smaller, specialized models.
Tokyo Gas i-net has implemented a solution that uses generative AI to capture and document the tacit knowledge of experienced project managers. The method ensures the organization retains critical decision-making criteria and experience even as employees leave the company.
Governance and policy7
Signal is blocking Gemini and Siri AI from reading your screen
Signal is restricting the ability to take screenshots on Android and iOS devices to prevent AI agents from reading users' messages and other data. The company argues that agentic AI capabilities in operating systems that analyze on-screen information undermine Signal's ability to independently guarantee privacy.
Governance Assess whether Signal should be included in internal guidelines for permitted AI usage, as the company's restrictions may impact the adoption of agentic AI tools.
Students sue Omnilert after AI mistook a bag of chips for a weapon (translated from de)
Four students are suing Omnilert, their school, and the police after an AI system incorrectly identified a bag of Doritos as a weapon. The incident resulted in the teenagers being handcuffed and searched. The case highlights the challenges of false positives in AI-based security monitoring.
Governance Assess the risk profile of AI-based security monitoring and ensure clear guidelines for human oversight in cases of false positives to avoid legal and ethical consequences.
US Department of Commerce approved AI chip exports to Chinese firms linked to the military (translated from ko)
The US Department of Commerce has approved the export of approximately 880,000 AI chips from Nvidia and AMD to Chinese companies. Over 30 percent of these chips were allocated to firms designated by the US Department of Defense as linked to the Chinese military, sparking significant controversy over the export review process.
Governance Review vendor risk management to ensure AI chips and models are not linked to military actors or high-risk jurisdictions.
Also today · 4
Eminem publisher Eight Mile Style has expanded its copyright infringement lawsuit against Suno by adding over 100 allegedly infringing tracks, including “Lose Yourself” and “Without Me.” The company has joined other entities such as Elvis Presley Enterprises and Reservoir in strengthening their legal claims against the generative AI service.
President Donald Trump has warned against building AI data centers outside the United States, threatening to impose a "tremendous tax" on companies that choose to locate their infrastructure abroad. This threat comes as several US states and cities implement restrictions on new applications for such facilities.
Also covered by: BBC - Technology
An analysis shows that Google's AI Overviews reduce clicks to external websites by 40 percent. The study also indicates that the feature often fails to provide users with useful or verified information.
The California Department of Justice has launched an investigation into Nvidia and former CEO Jensen Huang. The probe aims to determine if the company leaked advanced AI chips or confidential information to China and whether it adequately disclosed these risks to investors.
Media and content production3
Canva allows users to disable AI features (translated from sv)
Canva has introduced a new toggle that allows users to disable most of the platform's AI features, including Magic Write and Magic Edit. However, the company has excluded Magic Grab and the background removal tool, stating that these features do not generate new content.
Governance Evaluate whether our internal AI usage guidelines should include an option to disable generative features to enhance data control and compliance.
Also today · 2
A new survey from UMG indicates that over 80% of Americans want the option to disable AI-generated music. This contrasts with a recent survey from Deezer, where only 41% of listeners said they would prefer to filter out all AI music.
At the 47th American Film Market (AFM) in Los Angeles, Reverse Clock Entertainment and Zhaoyuan Culture Group presented the AI films "Effigy" and "Confirmation." The companies signed several cooperation agreements with international buyers for distribution and licensing. The film "Effigy" combines traditional Chinese paper-crafting with AI technology to tell a family story, receiving positive feedback from Hollywood attendees.
HR and customer support1
Workday’s Honeydew Deal Could Redefine AI Recruiting
Workday is transforming its recruiting solution from a traditional ATS into an AI-driven talent platform through acquisitions including Honeydew. Honeydew adds a semantic layer that standardizes definitions and metrics, enabling reliable analytics and agentic AI. This aims to help recruiting influence enterprise-wide workforce decisions.
Licences Evaluate whether Workday's new AI platform and Honeydew integration should be included in upcoming license negotiations or replace existing ATS licenses.
Other1
Mathematician claims to have solved Riemann hypothesis with AI (translated from no)
A mathematician claims to have solved the Riemann hypothesis, one of the biggest unsolved problems in mathematics, using artificial intelligence. A Norwegian professor calls it the biggest mathematical news in 200 years.
All 12 categories, every story today — nothing held back