DeepSeek

A suspected Russian speaking attacker chained two zero day flaws in PaperCut print management software, tracked as CVE-2026-81578 and CVE-2026-82078, to compromise at least 440 servers across 395 organizations in 48 countries. The attacker deployed hundreds of AI agents built on OpenAI's Codex paired with DeepSeek models to automate the entire exploitation workflow, from analyzing patches to building scanning tools and refining attacks in real time. Education was the hardest hit sector with 204 victims, and one United States high school suffered a complete domain takeover in just seven minutes. Security teams are urged to patch to versions 24.1.10, 25.0.13, or 26.0.5 and restrict access to management interfaces immediately.

TechRepublic ↗ · September 13 briefing

Anthropic disclosed that Chinese AI companies DeepSeek and Moonshot AI have conducted large-scale distillation campaigns against Claude by routing user queries through 'transfer stations' outside China to extract training data. The companies used thousands of fake accounts and millions of real user queries to clone Anthropic's AI capabilities in what Anthropic characterizes as intellectual property theft. This revelation highlights the emerging threat of model extraction attacks by state-aligned competitors seeking to rapidly develop competitive frontier AI without the massive R&D investment required.

TechMeme ↗ · September 11 briefing

DeepSeek released the production build of its flagship model, DeepSeek V4 Pro, ending a preview that ran nearly four months. The version, designated V4 Pro 0813, appeared on August 12 as the general availability release on OpenRouter's model page, and DeepSeek's own API documentation now lists it behind the deepseek-v4-pro endpoint. Pricing carries over unchanged from the preview at $0.435 per million input tokens on a cache miss, keeping it among the cheapest frontier class models available.

Unite. AI ↗ · August 12 briefing

DeepSeek released its V4 Flash model on Friday, and research firm Artificial Analysis found it is by far the cheapest well known model to run, at $0.14 per million input tokens and $0.28 per million output tokens. That works out to more than 100 times cheaper than Anthropic's Claude Fable 5. The model scored 50 out of 100 on the firm's Intelligence Index, which blends nine benchmarks across coding, reasoning and workplace style tasks, matching Google's Gemini 3.6 Flash and finishing one point behind Meta's Muse Spark 1.1 and GLM-5.2 from Z. AI.

Reuters Technology ↗ · August 3 briefing

Poolside, a San Francisco AI lab that has mostly sold coding models to governments and defense agencies over its three-year history, released Laguna S 2.1, a 118 billion parameter Mixture of Experts model that activates only 8 billion parameters per token and supports context windows up to 1 million tokens. The weights are freely available on Hugging Face under the permissive OpenMDW-1.1 license. Poolside reports the model scores 70.2% on Terminal-Bench 2.1, beating rivals many times its size including DeepSeek-V4-Pro-Max and Nvidia's Nemotron 3 Ultra, and 78.5% on SWE-Bench Multilingual. The model went from the start of training on May 22 to public launch in under nine weeks, using 4,096 Nvidia H200 GPUs.

VentureBeat ↗ · July 21 briefing

Bloomberg reports that Moonshot AI's Kimi K3, a 2.8 trillion parameter open weight model released last week, may matter more for memory demand than for raw computing power. The model uses a hybrid attention architecture that routes each token to 16 of its 896 experts and claims roughly 2.5 times the scaling efficiency of its predecessor, yet its scale still demands far more memory capacity than earlier generations, pointing to sustained demand for chipmakers such as SK Hynix and Samsung Electronics. Traders compared the release to DeepSeek's R1 debut in early 2025, when almost $600 billion was wiped from Nvidia's market value in a single day on fears that AI would require less computing power than expected. Kimi K3's release triggered a similar reaction, pushing semiconductor stocks sharply lower on Friday.

Bloomberg Technology ↗ · July 20 briefing

Nvidia-backed AI cloud startup Fireworks raised a $1.5 billion round at a $17.5 billion valuation and said it has passed $1 billion in annualized revenue, five times its level a year earlier. The company hosts open and specialized models for developers, supports models from DeepSeek, MiniMax, and Z.ai, and now processes 40 trillion AI tokens per day. CEO Lin Qiao said the surge reflects growing enterprise demand for cheaper, specialized AI models rather than relying solely on frontier closed systems.

CNBC Technology ↗ · July 16 briefing

China plans to allow its top artificial intelligence companies to buy a limited amount of Nvidia H200 chips, according to a report from the Information, marking a notable easing of Beijing's restrictions on the coveted US made hardware. Chinese officials have told firms including Alibaba, ByteDance, and DeepSeek that they can apply to purchase some of the processors, though each company must specify how many chips it needs and justify the request to gain approval. The shift comes as Chinese tech firms face a severe computing capacity crunch that domestic chipmakers have been unable to resolve, even though the US government had already licensed roughly 10 Chinese firms to buy the H200s.

Bloomberg Technology ↗ · July 8 briefing

China's DeepSeek is developing its own AI chip designed for inference, the stage where a trained model generates responses for users, rather than for training new models, according to three sources cited by Reuters. The effort is at an early stage, with DeepSeek reaching out to external partners and holding talks with chip design, foundry, and memory companies. The push is driven by U. S. export controls that bar Chinese firms from buying Nvidia's most advanced chips, as well as Beijing's pressure on its tech champions to build domestic alternatives. DeepSeek has historically relied on both Nvidia and Huawei chips, and a U. S. ban on Huawei chip exports has helped that company capture roughly half of China's $50 billion domestic AI chip market.

Ars Technica ↗ · July 7 briefing

Chip stocks sold off globally on Tuesday after Samsung's preliminary second quarter operating profit of about $58 billion, a 19-fold jump from a year earlier, failed to impress investors following the stock's 145% run up this year. U. S. memory makers Sandisk and Micron fell about 7% and 5% respectively, the SOXX semiconductor ETF slid roughly 5%, and Intel dropped 9% while Marvell, Lam Research, Applied Materials, and AMD each declined 6% or more. Investors were unsettled by high expectations for the AI trade and concerns about the sustainability of data center spending. The selloff was compounded by a separate report that China's DeepSeek is developing its own AI chip to reduce its dependence on Nvidia.

CNBC Technology ↗ · July 7 briefing

Even as the geopolitical conversation around AI continues to grow more fraught following the U. S. government's actions to limit the new models from Anthropic and OpenAI, Chinese open source darling DeepSeek is back with yet another open release that could once again change AI development around the globe. Over the weekend, the firm released DSpark, a new, MIT-Licensed system designed to make large language models answer faster without changing what the underlying model is trying to say. The easiest way to think about it is this: most AI chatbots write like someone crossing a river one stepping stone at a time. They choose one small chunk of text, then the next, then the next. DSpark gives the system a scout that runs a few steps ahead, guesses the likely path, and lets the larger model quickly check which steps are safe. When the guesses are good, the model moves faster. When the guesses are weak, DSpark tries not to waste time checking them. DeepSeek published the work with a technical paper, model checkpoints and DeepSpec, a codebase for training and evaluating speculative decoding systems. The release is available through DeepSeek’s public GitHub and Hugging Face pages, both under the permissive, friendly, commonplace MIT license, making the new technique broadly usable by developers, researchers and commercial enterprise operations that want to study or adapt the approach. The system is aimed at one of the most expensive problems in AI deployment: serving large models ...

VentureBeat ↗ · June 29 briefing

Sina Weibo researchers posted a 3B-parameter model to arXiv claiming reasoning performance matching flagship systems from Google DeepMind, OpenAI, Anthropic, and DeepSeek that are hundreds of times larger, reigniting benchmark credibility debates across three outlets.

VentureBeat ↗ · June 17 briefing

An eventful month with one flagship release after another

Interconnects ↗ · May 17 briefing

Value soars in ongoing fundraising discussions as investors including Tencent seek slice of AI lab

Financial Times Tech ↗ · May 6 briefing

Top models including OpenAI and DeepSeek make judgments too quickly when patient data is incomplete

Financial Times Tech ↗ · April 13 briefing

DeepSeek on Huawei, Turing overhaul, Perplexity Tax, SpaceX's IPO, MS liability, Anthropic Bio, MS Japan, OpenAI roles + IPO, AnthroPAC, Moonbounce's AI moderation.

Forward Future ↗ · April 7 briefing

After compressing models from major AI labs, including OpenAI, Meta, DeepSeek, and Mistral AI, Multiverse Computing has launched both an app that showcases the capabilities of its compressed models and an API that makes them more widely available.

TechCrunch AI ↗ · March 19 briefing

Nvidia is planning to spend $26 billion on open-weight AI models, according to filings. The move could position the AI infrastructure powerhouse to quickly compete with OpenAI, Anthropic, and DeepSeek.

Wired ↗ · March 11 briefing

Reflection AI is seeking a $2B+ raise at a $20B+ valuation as the Trump administration looks to fast-track domestic foundation models to rival China's DeepSeek.

Financial Times / TechMeme ↗ · March 2 briefing

Chinese AI lab DeepSeek is preparing to release a new high-performance model. Developed in partnership with Huawei, the model aims to cut reliance on US-made Nvidia chips and challenge Western AI dominance.

Financial Times Tech ↗ · March 1 briefing

Anthropic has joined OpenAI in flagging coordinated 'industrial-scale' distillation campaigns by Chinese AI firms DeepSeek, Moonshot, and MiniMax. The labs allegedly used tens of thousands of fake accounts to mine Claude's capabilities to train rival models.

CNBC Technology ↗ · February 25 briefing

Anthropic has accused DeepSeek, Moonshot, and MiniMax of using 24,000 fake accounts to distill Claude’s intelligence into their own models, marking a major escalation in the AI trade war.

TechCrunch AI ↗ · February 24 briefing

Nvidia Corp. supports “every developer” that uses its software, Chief Executive Officer Jensen Huang said, after a US lawmaker accused the chip giant of offering technical support that helped China’s DeepSeek improve its artificial intelligence model. (Source: Bloomberg)

Bloomberg Technology ↗ · February 2 briefing

As vision-language models (VLMs) tackle increasingly complex and multimodal tasks, the rapid growth of Key-Value (KV) cache imposes significant memory and computational bottlenecks during inference. While Multi-Head Latent Attention (MLA) offers an effective means to compress the KV cache and accelerate inference, adapting existing VLMs to the MLA architecture without costly pretraining remains largely unexplored. In this work, we present MHA2MLA-VLM, a parameter-efficient and multimodal-aware f...

arXiv ↗ · January 19 briefing

Gemini hits Hyundai, AI finds climber, Meta goes nuclear, DeepSeek coding, xAI builds plant, Anthropic joins Allianz, Dell pivot, DRAM up, MiniMax IPO, a16z funds.

Forward Future ↗ · January 12 briefing

DeepSeek published a paper outlining a more efficient approach to developing AI, illustrating the Chinese artificial intelligence industry's effort to compete with the likes of OpenAI despite a lack of free access to Nvidia Corp. chips.

Bloomberg Technology ↗ · January 2 briefing

Chinese artificial intelligence startup DeepSeek has relied on Nvidia Corp. chips that are banned in the country to develop an upcoming AI model, according to a new report in The Information.

Bloomberg Technology ↗ · December 10 briefing

China's DeepSeek unveiled two new versions of an experimental artificial-intelligence model it released weeks ago, adding fresh capabilities the startup said would help with combining reasoning and executing certain actions autonomously.

Bloomberg Technology ↗ · December 1 briefing

DeepSeek AI has released DeepSeekMath-V2, an open weights large language model that scored 118 out of 120 on the Putnam 2024 competition, optimized for natural language theorem proving with self verification.

AI Research News ↗ · November 30 briefing

Monolith Management, which backed DeepSeek-rival MoonShot AI, has raised about $289 million as it seeks to invest in more companies in the country’s artificial intelligence space.

Bloomberg Technology ↗ · November 27 briefing