EP 626 : Anthropic Threat Report Reveals Claude Misuse & DeepSeek’s Cheap Flash Model – AI News

Feeling uneasy after yesterday’s AI apocalypse headlines? This episode flips the script with Anthropic’s eye‑opening Threat Report, exposing the wildest ways bad actors tried to weaponize Claude over the past eight months. We walk you through seven Chinese labs—Alibaba, DeepSeek, Moonshot, Xiaomi and others—caught distilling Claude, some even re‑branding it as their own. From bioweapon‑flagged biology experiments and a Yemeni rocket‑guidance hack to a Mali spy‑agency phone‑surveillance system, Anthropic shut down the attempts and now shares the gritty details.We then shift to the price war heating up in China. DeepSeek just launched V4.1‑Flash, an open‑weight model that undercuts Claude Opus 5 and even GPT‑5.6 on agentic, coding and cyber benchmarks while costing as little as $0.15 per million tokens. Its MIT‑licensed weights are live on Hugging Face, promising a new volume‑first playbook for developers.Our rapid‑fire AI news roundup also covers Cognition’s SWE‑2 coding model, Andrew Tulloch’s move from Meta to Anthropic, OpenAI’s denial of a NYT claim about Codex‑driven Navier‑Stokes proof, UMG’s ElevenLabs licensing deal, Tavus’s hyper‑realistic Phoenix 4.5 human avatar, and the latest ChatGPT Work data plugin that pulls company data straight into a chat. Plus a quick glance at the hottest tools you can start using today.Tune in for the full story, share with your tech‑savvy friends, and hit subscribe so you never miss a beat in the fast‑moving AI landscape.Tools mentioned: DeepSeek V4.1‑Flash, Cognition SWE‑2, OpenAI GPT‑Live‑1, Krea Agents, Figr, Typecast, NxCode.