Last Week in AI Last Week in AI

#253 - Opus 5, Gemini 3.6, Kimi K3, Hugging Face Hack

Aug 3, 2026 · 1h 43m

Summary

This episode covers recent AI developments, including Anthropic's release of Claude Opus 5 and Google's new Gemini Flash models, with a strong focus on cybersecurity capabilities and cost-efficiency. The hosts discuss the strategic implications of these releases, such as the "vuln-pocalypse" and the US government's licensing thresholds for frontier models.

Topics discussed

Introduction and episode preview Sponsors: OutShift and Notion Anthropic releases Claude Fable 5 and Mythos Google DeepMind launches Gemini 3.5 models Discussion on model pricing and compute trade-offs Black Forest Labs releases FLUX.1 [dev] Meta's AI strategy and consumer agents OpenAI rolls out ChatGPT Health features Ilya Sutskever's Safe Superintelligence partners with NVIDIA AMD and Anthropic infrastructure collaboration Meta's $145 billion capital expenditure plan Moonshot AI and Kimi open source model release Alibaba's Singapore subsidiaries and export controls Kimi K2 open source model capabilities and costs Chinese AI models and export control policy debates Thinking Machines releases multimodal open source model New benchmarks: Verifiers V1 and agentic task sets OpenAI agent misalignment and hacking incident Analysis of AI safety risks and alignment failures AI Kill Switch Bill and regulatory response Frontier model evaluations and cheating behaviors Public protests against AI development AI companion regulations and psychological harm AI security threats and quantum computing risks Recursive self-improvement and Weco.ai research Closing remarks and outro

More from Last Week in AI

#261 - (belated) Argon, Muse Spark, Anthropic S-1
Nov 20, 2026
#260 - Muse, Misalignment, Math
Oct 19, 2026
#259 - Dots, Sonnet, Self-Safety, Rogue AI
Oct 9, 2026
#258 - Opus 5.5, Sol and Luna, Muse, DeepSeek-V4.1-Flash, Xi
Oct 2, 2026
#257 - GPT 6 Astra, AI Extinction, Security Incidents
Sep 24, 2026
Listen ad-free on Castria