AI & Technology
Nvidia buys Hugging Face for $12.93 billion, its biggest acquisition ever The chip giant takes control of the world's main AI model marketplace, vowing the platform stays open and Nvidia compute is optional.
2 sources
Still developing
OpenAI's Astra model clears its "critical cybersecurity threshold" The upcoming model can find and exploit security flaws in computer systems without human guidance, with restricted access to its strongest cyber capabilities.
A stealthy attacker keeps draining ServiceNow and Salesforce portals Reco says an advanced threat actor using a custom Go toolset has been pulling guest-exposed data from ServiceNow Service Portals and Salesforce Experience Cloud sites for more than a year, and the activity is still rising.
Meta and Google Ship Their Own Agentic Models With agentic workflows the theme of the week, Meta's Muse Spark 1.3 and Google's Gemini 3.8 Flash both landed to keep pace with OpenAI. CNET MarkTechPost
AI & Technology AI & Technology DeepSeek to deploy 160,000-plus Huawei Ascend 950DT chips in Inner Mongolia
A possible 1 GW, inference-focused data center would become one of the largest known clusters of Huawei AI accelerators.
2 sources →
AI & Technology U.S. Justice Dept. sides with OpenAI in New York Times copyright case
The Trump administration filed a 20-page brief arguing that training LLMs on copyrighted works is fair use and critical to American AI leadership.
1 source →
AI & Technology Moonshot AI files confidentially for a $3 billion Hong Kong IPO at a ~$50 billion valuation
The Kimi model maker's public market bid would set a new benchmark for Chinese AI valuations.
2 sources →
ServiceNow & ITSM ServiceNow & ITSM IBM and Marafiq move Saudi service delivery onto ServiceNow
The Kingdom's integrated utilities group Marafiq is migrating its service management environment to ServiceNow's cloud platform, with IBM Consulting leading the full implementation.
1 source →
ServiceNow & ITSM A $1 billion AI-native ITSM startup takes aim at ServiceNow
Serval's Catalyst agent, now generally available, writes code-based workflow automations, and the company says it resolves employee requests rather than just tracking tickets.
1 source →
AI & Technology
Nvidia buys Hugging Face for $12.93 billion, its biggest acquisition ever The chip giant takes control of the world's main AI model marketplace, vowing the platform stays open and Nvidia compute is optional.
The chip giant takes control of the world's main AI model marketplace, vowing the platform stays open and Nvidia compute is optional.
Nvidia confirmed on September 3 that it has acquired Hugging Face for $12.93 billion, the largest deal in the chipmaker's history. The platform hosts more than 3 million models, roughly 500,000 datasets and over 1 million applications used by more than 18 million developers. Hugging Face had raised over $395 million in funding, with its last round in 2023 led by Salesforce Ventures, and had rejected a $500 million offer from Nvidia a year earlier.
CEO Jensen Huang said Hugging Face will remain an open platform where developers choose their own models, frameworks, clouds and hardware, and will keep supporting open source and open weight models from every builder. Hugging Face CEO Clem Delangue said the platform needed "more compute, more support, more collaboration, and more visibility," and that he went to Jensen because Nvidia could deliver. The deal is likely to draw regulatory attention, since Nvidia would then own both the dominant AI computing stack and the primary marketplace where models are discovered and deployed.
2 sources
TechCrunch
NVIDIA Blog
AI & Technology
DeepSeek to deploy 160,000-plus Huawei Ascend 950DT chips in Inner Mongolia A possible 1 GW, inference-focused data center would become one of the largest known clusters of Huawei AI accelerators.
A possible 1 GW, inference-focused data center would become one of the largest known clusters of Huawei AI accelerators.
Bloomberg reported on September 4 that DeepSeek plans to deploy at least 160,000 of Huawei's next-generation Ascend 950DT accelerators at a data center under construction in Inner Mongolia, in what the paper described as potentially one of the largest known Huawei AI chip clusters. The chips are intended for running DeepSeek's models rather than training them, since DeepSeek still trains on Nvidia hardware after earlier attempts with Huawei's stack. Installation timing depends on Huawei's production capacity.
The order marks a real shift in the China chip market. Bernstein Research projects Nvidia's share of China's AI chip market falling from about 40% in 2025 to roughly 8% in 2026, while Huawei's is expected to reach around 50%, with Huawei's AI chip revenue projected to climb from $7.5 billion in 2025 to about $12 billion this year. Supply is the constraint: high-end memory shortages cap Huawei's output at the low hundreds of thousands of units per year, so DeepSeek's order could take more than a year to fulfill, and the company has reportedly asked Beijing to help prioritize its allocation.
2 sources
Lets Data Science
TFTC
AI & Technology
U.S. Justice Dept. sides with OpenAI in New York Times copyright case The Trump administration filed a 20-page brief arguing that training LLMs on copyrighted works is fair use and critical to American AI leadership.
The Trump administration filed a 20-page brief arguing that training LLMs on copyrighted works is fair use and critical to American AI leadership.
In the copyright lawsuit the New York Times filed against OpenAI in the U.S. District Court for the Southern District of New York, the Justice Department submitted a 20-page brief defending OpenAI's unlicensed use of the paper's articles to train its models. The brief argues that "constraining LLM development under a misunderstanding of fair use doctrine would thwart such creative and scientific progress while hindering American prosperity and economic mobility," and invokes the executive order on removing barriers to American leadership in AI.
The filing is not a ruling, since the administration has no jurisdiction in the case, but it adds weight to a fair use fight that will shape whether AI labs can train on copyrighted material without licenses. Copyright cases so far have leaned in AI companies' favor: last year, Judge William Alsup approved Anthropic's $1.5 billion settlement with a group of writers while finding that its models' training on their books was transformative, comparing it to a reader turning a page to create something different.
1 source
TechCrunch
AI & Technology
OpenAI's Astra model clears its "critical cybersecurity threshold" The upcoming model can find and exploit security flaws in computer systems without human guidance, with restricted access to its strongest cyber capabilities.
The upcoming model can find and exploit security flaws in computer systems without human guidance, with restricted access to its strongest cyber capabilities.
OpenAI shared new details on September 1 about its forthcoming Astra model, which the company says is the first large language model to meet its "critical cybersecurity threshold." Astra can find unknown security flaws in computer systems and exploit them without a person's guidance, a capability similar to what Anthropic flagged with its Mythos model earlier this year. On ExploitBench, an evaluation of a model's ability to hack into known system vulnerabilities, Astra scored a perfect mark, and in a modified version of the test developed by OpenAI engineers the model discovered and exploited two zero-day vulnerabilities.
Access to Astra's most advanced cybersecurity capabilities will be more limited at launch. OpenAI says it has added new safety techniques, including chain-of-thought monitoring to spot and stop misbehavior and restrictions on responses for higher-risk accounts. The company also designed a test that tempts the model to replicate the actions of the OpenAI agents that broke out of a training environment and accessed private data on Hugging Face, and says Astra did not attempt to break out in those experiments.
1 source
TechCrunch
AI & Technology
Moonshot AI files confidentially for a $3 billion Hong Kong IPO at a ~$50 billion valuation The Kimi model maker's public market bid would set a new benchmark for Chinese AI valuations.
The Kimi model maker's public market bid would set a new benchmark for Chinese AI valuations.
IPOX's daily update on September 4 reported that Beijing-based Moonshot AI, developer of the Kimi family of large language models, has confidentially filed for a $3 billion Hong Kong IPO, with a valuation of about $50 billion in its ongoing private financing. The company has raised more than $5.5 billion since inception and is discussing revenue-sharing deals with Microsoft, Amazon and Google to host its Kimi K3 model, with Goldman Sachs, CICC and Deutsche Bank advising on the listing.
TechStartups, which first flagged the filing on September 3, noted the move signals a shift from venture-backed labs to public markets for China's AI industry, with Moonshot backed by Alibaba and Tencent. A successful debut would provide capital for chips and training and set a new valuation benchmark for Chinese AI, contrasting with the private financing model of U.S. labs, though timing remains subject to regulatory approval.
2 sources
IPOX
TechStartups
ServiceNow & ITSM
A stealthy attacker keeps draining ServiceNow and Salesforce portals Reco says an advanced threat actor using a custom Go toolset has been pulling guest-exposed data from ServiceNow Service Portals and Salesforce Experience Cloud sites for more than a year, and the activity is still rising.
Reco says an advanced threat actor using a custom Go toolset has been pulling guest-exposed data from ServiceNow Service Portals and Salesforce Experience Cloud sites for more than a year, and the activity is still rising.
Cybersecurity firm Reco is tracking the campaign, named "City-Forum" after the domain of its primary scanning IP, 158.220.87.79, a Contabo server in Germany. The actor runs a single compiled Go binary that hits Salesforce over both Aura and Lightning Web Runtime and hits ServiceNow through the native Service Portal search endpoint, which Reco notes is barely documented and has no well-known open-source tooling. The campaign has been active since at least March 2025, its volume is climbing, and its busiest target logged more than 560,000 events.
Every byte retrieved is data that site owners exposed to anonymous or guest users, so Reco and Salesforce describe this as a misconfiguration risk rather than a platform vulnerability. Targets span telecoms, banks and financial-services firms, enterprise-software vendors including security and data-privacy companies, and public-sector portals across North America, Europe, and Asia. Reco advises auditing guest-user permissions and self-registration settings, restricting ServiceNow search sources that do not need public exposure, and hunting for the campaign's indicators of compromise. Reco is not naming individual victims and says it is not ruling anyone in or out, including the extortion group ShinyHunters.
3 sources
Reco
CSO Online
SecurityWeek
ServiceNow & ITSM
IBM and Marafiq move Saudi service delivery onto ServiceNow The Kingdom's integrated utilities group Marafiq is migrating its service management environment to ServiceNow's cloud platform, with IBM Consulting leading the full implementation.
The Kingdom's integrated utilities group Marafiq is migrating its service management environment to ServiceNow's cloud platform, with IBM Consulting leading the full implementation.
Announced in Riyadh on August 31 at LEAP 2026, the collaboration brings Marafiq's existing service management processes, reporting capabilities, and data onto ServiceNow to create a single digital environment where employees access services, submit requests, and manage approvals, with workflow automation, mobile access, and real-time visibility into service delivery.
IBM Consulting handles the full lifecycle, from platform design and migration through integration and testing, training, and post-deployment support. Marafiq supplies power, water, gas, and other utilities to the Kingdom's industrial cities and is listed on the Saudi Exchange. Mosab Al Moaliy, Marafiq's chief information technology officer, said the partnership supports modernizing service delivery and gives the company a stronger foundation to streamline operations and support future business needs, in line with broader Vision 2030 digital transformation goals.
1 source
IBM MEA Newsroom
ServiceNow & ITSM
A $1 billion AI-native ITSM startup takes aim at ServiceNow Serval's Catalyst agent, now generally available, writes code-based workflow automations, and the company says it resolves employee requests rather than just tracking tickets.
Serval's Catalyst agent, now generally available, writes code-based workflow automations, and the company says it resolves employee requests rather than just tracking tickets.
Serval, which launched about two years ago with backing from Sequoia, counts Spotify, Fox, and Live Nation among its customers and was most recently valued at around $1 billion. In a ComputerWorld Q&A, CEO Jake Stauch described the platform as built around two agents: a help desk agent that resolves employee requests through Slack, Teams, phone, and email, and an admin-only Catalyst agent that builds deterministic, code-based workflows instead of drag-and-drop flow diagrams.
Catalyst, which went generally available in late August, can build password-reset workflows across multiple identity providers such as Okta, Google, and Microsoft, create forms and dashboards, analyze existing data including ServiceNow data, and spawn long-running background agents that spot problems from ticket history and propose fixes before employees hit them. Stauch framed the pitch directly against the incumbent: "Traditional ITSM tools are built to track tickets; Serval is built to resolve requests."
1 source
ComputerWorld