South Korea's data protection authority said on Thursday that Chinese artificial intelligence startup DeepSeek transferred user information and prompts without permission when the service was still available for download in the country's app market.
https://tech.yahoo.com/articles/south-korea-agency-says-deepseek-032105626.html
The incident should serve as a critical wake-up call. The stakes are simply too high to treat #AI #security as an afterthought — especially when the Dark Web stands ready to capitalize on every #vulnerability.
https://www.darkreading.com/cyberattacks-data-breaches/deepseek-breach-opens-floodgates-dark-web
#DeepSeek
"#DeepSeek, and the #opensource #AIecosystem surrounding it, has rapidly evolved from a brief snapshot of technological brilliance into something much bigger—and much harder to stop. Tens of thousands of developers, from seasoned #researchers to passionate #hobbyists, are now working on enhancing, tuning, and extending these open source models in ways no centralized entity could manage alone."
https://www.infoworld.com/article/3960764/deepseeks-open-source-movement.html
Another reminder of the virtues of running #LLM locally e.g. via #Ollama
DeepSeek Breach Opens Floodgates to Dark Web
> [Researchers discovered] a publicly accessible .. database belonging to #DeepSeek, containing more than a million lines of log streams with highly sensitive information. This exposed data included chat history, #API keys ...
https://www.darkreading.com/cyberattacks-data-breaches/deepseek-breach-opens-floodgates-dark-web
The wait is over! #76: Shiny Objects that People Like to Chase is out!
@kito99, @dhinojosa, and Ian start out the new year with #WebAssembly, #SemanticWeb, and #AI guru @bsletten. They talk about #WebAssembly, #LLMs, edge computing, and open source hardware. The crew also discusses how theory of mind relates to #AGI, #DeepSeek, #OpenWeb, #Fediverse, #ActivityPub, Interplanetary File System (#ifs), and more. https://www.pubhouse.net/podcast/title-shiny-objects-that-people-like-to-chase/
DeepSeek entzaubert: Stärken und Schwächen
In den vergangenen Wochen habe ich das lokal auf meinem Rechner laufende KI-Sprachmodell "DeepSeek-R1-Distill-Qwen" immer wieder für unterschiedliche Anwendungsfälle (z.B. Programmierung, Textzusammenfassung, Übersetzung, etc.) ausgetestet.Welche Stärken und Schwächen es hat und wofür es sich in der Praxis wirklich nutzen läßt, erfahrt Ihr in diesem Video :-)
#AI #DeepSeek #KI #KünstlicheIntelligenz
https://tube.tchncs.de/w/6GyWqLXTXmfbV8TKARpwCr
https://youtu.be/2tOvhuAaLKQ
@gerrymcgovern #Deepseek appears to be showing that AI may not be as processing-intensive as companies like #Amazon and #microsoft believed.
Researchers have uncovered a new supply chain attack called #Slopsquatting where threat actors exploit hallucinated, non-existent package names generated by #AI coding tools like #GPT4 and #CodeLlama
These believable yet fake packages (amounting to 19.7% or 205,000 packages), recommended in test samples were found to be fakes., can be registered by attackers to distribute malicious code.
Open-source models -- like #DeepSeek and #WizardCoder -- hallucinated more frequently, at 21.7% on average, compared to the commercial ones (5.2%) like GPT 4.
We Have a Package for You! A Comprehensive Analysis of Package Hallucinations
by Code Generating LLMs (PDF) https://arxiv.org/pdf/2406.10279
Google argues its survival is crucial for US national security and competition with China, as it faces potential breakup amid DOJ's antitrust hearings
#YonhapInfomax #Google #DOJ #AntitrustHearing #AICompetition #DeepSeek #Economics #FinancialMarkets #Banking #Securities #Bonds #StockMarket
https://en.infomaxai.com/news/articleView.html?idxno=59670
@mrola Tengo meses usando #DeepSeek para generar código, y es mucho más eficiente, limpio y bien documentado, que otras herramientas similares. He podido generar scripts incluso para herramientas en iOS, como #Pythonista o #Scriptable con facilidad.
"As the launch of #DeepSeek’s next-generation model is awaited, another previously little-known Chinese startup, Butterfly Effect, rolled out #Manus, an invite-only AI agent, last month. It claimed Manus is the world’s first #generalAI agent, which can perform tasks with a degree of autonomy that current AI models lack."
https://restofworld.org/2025/china-ai-boom-chip-ban-deepseek/
#EstadosUnidos quiere prohibir #DeepSeek
- mpr21
https://mpr21.info/estados-unidos-quiere-prohibir-deepseek/
A new U.S. House report on DeepSeek highlights how one Chinese AI model may be quietly reshaping global AI strategy — and risking American data privacy.
The House Select Committee on the CCP has released findings on DeepSeek’s R1 model, revealing:
- $420M in funding from High-Flyer Quant, a Chinese trading firm
- Access to 10,000+ NVIDIA A100 chips via the Firefly supercomputing infrastructure
- Ties to China's surveillance ecosystem, including China Mobile
- Allegations of illegal training data use and export control circumvention
- App behavior that mimics spyware: collecting device IDs, typing cadence, and chat history
Lawmakers warn that DeepSeek:
- Functions as an open-source intelligence asset for China
- Circumvented guardrails from U.S. AI companies to accelerate its own development
- Operates under a tightly controlled tech ecosystem with deep state-linked partnerships
An OpenAI exec told the committee that DeepSeek “circumvented guardrails to extract reasoning outputs,” accelerating their model using techniques like distillation — potentially copying U.S. tech at lower cost.
Even more concerning:
- User data is routed via infrastructure tied to China Mobile
- DeepSeek does not encrypt much of its traffic
- It censors content critical of the Chinese government
What this means:
- Export controls alone aren’t enough — the U.S. must improve early threat tracking
- Agencies should restrict procurement and usage of Chinese AI models
- More visibility and scrutiny are needed around AI supply chains and infrastructure
At Efani, we believe real AI security starts with understanding who’s behind the tools we use — and where our data ends up. This report is a wake-up call for all of us building or relying on AI systems today.
Do "reasoning" #LLMs like #DeepSeek's truly deliberate?
Wang et al. found such #AI systems exhibited cognitive biases.
And injecting phrases like "wait, let me think about it" may have exacerbated one bias!
They dub this "superficial reflection bias".
https://doi.org/10.48550/arXiv.2504.09946
I treat #AI like special needs students. Some can see. Some can't. Some can remember. Some can't. Some can fold proteins. Most can't. US #AI won't talk about politics. But other countries #AI will. #DeepSeek is Chinese. #FromTheMouthsOfBabes
@NORADCommand I recently read where the US Military is going to spend on the development of the @USNavyResearch
I read a study decades ago that said that submarines were needed but ships were vulnerable. I just had one hell of a conversation with #Deepseek #AI Thought I'd share
