Google DeepMind has released Gemini 3.5 Flash Cyber, a model designed to "automatically builds exploit code" for security testing. The launch comes as national security experts warn that the spread of powerful open-source models could soon unleash "disruptive cyber attacks at massive scale."
The tape Last Week in AI · Aug 3
“And then there is of course, the cybersecurity focused Gemini 3.5 Flash Cyber, which is integrated into this code mentor agent that automatically builds exploit code to verify vulnerabilities in sandboxed environments and then generates patches, which they say has found problems in complex real pieces of software such as V8 JavaScript engine.”
16:44 / --:--
The Artificial Intelligence Show · #229: Q3 Trends Briefing - The Pope's AI Encyclical, AI Agents Hack Hugging Face, Fable 5 vs. Washington, and the Battle Over Open WeightsAug 6
Despite public narratives of AI augmenting jobs, private conversations with executives reveal plans for efficiency-driven layoffs. The claim comes as companies like Coinbase and Cloudflare openly cite AI while cutting up to 20% of their workforce.
How I AI · Build an AI code review bot in 30 minutes with Vercel EveAug 5
A new wave of AI coding assistants is starting to autonomously review and approve software changes, with early adopters like Intercom reporting a 2-3x increase in pull request throughput. The bots score code for risk, auto-approving safe changes and escalating riskier ones to human engineers.