Deepseek is a topic tracked in our intelligence system with 5 linked articles.
DHS has ordered all federal civilian agencies to strip DeepSeek products — including third-party offerings built on DeepSeek implementations — from their IT systems and prohibited future purchases from the Chinese AI developer or its affiliates.
The Department of Homeland Security will ban DeepSeek's AI from federal civilian agency systems over national security concerns, including data transfers to China-based servers, potential malware, and risk to federal IT infrastructure.
Chinese hacker group allegedly commands DeepSeek via Telegram to orchestrate autonomous cyberattacks.
Mistral is leveraging European regulatory dynamics and a shift toward open-weight/open-source AI to gain traction against US peers, backed by near-$2B funding at a multi-trillion-valuation trajectory and rising revenue driven by government and enterprise partnerships.
AirLLM claims 70B model inference on a 4GB GPU via sparse MoE streaming, with ongoing updates (e.g., Kimi K3 support) and multiple model families showing low VRAM footprintsisd; it also documents compression options to further cut memory and speed requirements.
ICML-backed research argues LLMs have a fundamental flaw that makes full security unattainable via chain-of-thought forgery, with broad safety and regulatory implications.
Subscribe for real-time topic updates and unlimited access to our intelligence platform.