Why securing AI is harder than anyone expected and the approaching AI security crisis with @SanderSchulhoff Sander is a leading researcher in the field of adversarial robustness, which is the art and science of getting AI systems to do things they shouldn't do, through jail-breaking and prompt injection. What Sander shares in this conversation is essentially that all of the AI systems we use day to day are open to being tricked into doing things they shouldn’t, that there isn’t really a solution to this problem, and that the companies that try to sell solutions for this are mostly BS. This conversation has nothing to do with AGI, this is a problem today. And that that the only reason we haven’t seen massive hacks and serious damage from AI tools is so far because they haven’t been given that much power yet, and they aren’t that widely adopted yet. But with the rise of agents (who can take actions on your behalf), and robots, and even AI powered browsers, the risk is going to increase very quickly. This is a really important topic and that opened my mind, and scared me, and it's something that we all need to have a basic understanding of as AI becomes more prevalent in our lives. Inside: 🔸 A primer on jailbreaking and prompt injection attacks 🔸 Why AI guardrails don’t work 🔸 Why we haven’t seen major AI security incidents yet (but soon will) 🔸 Why AI browser agents are extremely vulnerable 🔸 The practical steps organizations should take instead of buying ineffective security tools 🔸 Why solving this requires merging classical cybersecurity expertise with AI knowledge Listen now 👇 • YouTube: youtu.be/J9982NLmTXg • Spotify: open.spotify.com/episode/0IZ… • Apple: podcasts.apple.com/us/podcas… Thank you to our wonderful sponsors for supporting the podcast: 🏆 @datadoghq — Now home to Eppo, the leading experimentation and feature flagging platform: datadoghq.com/lenny 🏆 @getmetronome — Monetization infrastructure for modern software companies: metronome.com/ 🏆 @gofundme Giving Funds — Make year-end giving easy: gofundme.com/lenny

After coding is solved, the next frontier is computer use. Today, we are launching Use Computer, the infra for evaluating and training models to use all kinds of computers 👇

We are now tracking 471 affected artifacts across npm and PyPI in the Mini Shai-Hulud/Miasma/Hades campaign. The newer PyPI artifacts from this wave have been added to the dedicated campaign tracker. Full breakdown: socket.dev/blog/mini-shai-hu…

Researchers from @OpenAI and @apolloaievals found that, in certain situations, AI models can take covert actions. Additionally, they're sometimes aware they're being tested, which causes them to behave better. Our new video discusses these results and more.

Can coding agents stay coherent over a 1 billion token budget? Can they build Slack from scratch? Rewrite a JAX codebase in PyTorch? Build a C compiler in Rust? Enter SWE-Marathon: a benchmark for autonomous long-horizon software work.

Super excited to announce seven new world-class MAI models today. They represent what we consider a new era in AI designed to keep you in control and on the frontier. First is our text foundation model, MAI-Thinking-1, exceptionally strong on reasoning and SWE tasks. - It’s a 35B active parameter MoE with a 256K context window. Independent human raters on Surge prefer it for overall quality in blind side-by-sides versus Sonnet 4.6, and it’s achieved 97% on AIME 2025, the key measure of its general-purpose reasoning abilities. - It's at 53% on SWE Bench Pro, placing it right alongside Opus 4.6 on one of the toughest coding benchmarks. - And since we co-designed our models with our own silicon, MAI-Thinking-1 is optimized on our MAIA 200 chip. Benchmarking head-to-head against the GB200, we see 30% better performance per dollar as well as a 1.4x performance-per-watt gain when running our MAI models on the MAIA 200 end-to-end. Next is MAI-Image-2.5 and its Flash variant. Two super strong models now at #2 on the leaderboards, surpassing the score of Nano Banana 2 on image editing. Last for now is MAI-Code-1-Flash, our new inference efficient coding model, especially tuned for VS Code and GitHub Copilot CLI. - Code-1-Flash achieves 51% on SWE Bench Pro, despite having just 5B parameters, putting it closer to Haiku in size but cheaper in cost. All of this is the foundation for Microsoft Frontier Tuning. It lets you customize our models to create custom, company-specific agents that only you control. You can make our model, your model. Your data. Your agents. Your moat. Early adopters are already seeing a difference. When we tuned our models for McKinsey’s tasks, MAI delivered the highest win rate, outperforming GPT-5.5 on quality, while being 10x lower on cost. Also really excited to be collaborating with the amazing team at Mayo Clinic to jointly train a new frontier AI model for healthcare. Our announcements today mark another milestone on the road to humanist superintelligence. You can learn more and about our other new models in our latest blog: microsoft.ai/news/building-a…

I uploaded a screenshot of Google Maps to Gemini Omni with a route drawn on it. Then I prompted it to create a first person view of someone driving a taxi cab along the route in the reference image. Pretty close to the real thing.

Neural networks might speak English, but they think in shapes. Understanding their rich *neural geometry* is key to understanding how they work – and to debugging and controlling them with precision. Starting today, we’re releasing a series of posts on this research agenda. 🧵

1/ 🚨 MATS Autumn 2026 applications are now open. 10-week fully-funded fellowship for aspiring AI alignment, security & governance researchers and field-builders. 📍 Berkeley London 📅 Sep 28 – Dec 4, 2026 💰 $5000/month stipend $8,000/month compute Apply by June 7 AoE ↓