#AppSec 1⃣ PromptJacking: Critical RCEs in Claude Desktop That Turn Questions Into Exploits koi.ai/blog/promptjacking-th… // RCE vulnerabilities in three extensions that were written, published, and promoted by Anthropic themselves - the Chrome, iMessage, and Apple Notes connectors 2⃣ Code execution via prompt injection in GitHub Copilot Chat jro.sg/CVEs/copilot // Since v0.32.2, editing requires user approval

Implementing the Etherhiding technique medium.com/@onhexgroup/imple…

remember ( piped.video/watch?v=ih8c1yT2… ) that in learning , the past ( piped.video/watch?v=iS2ET12C… ) don't exist and there's no limit ( piped.video/watch?v=lEia7Qmb… ) and no return ( piped.video/watch?v=VFk8_hvU… ) , just one last thing ( piped.video/watch?v=WdH1cCKm… ) : eternity ( piped.video/watch?v=y-XLABr3… )

To advance foundational AI, we must solve continual learning and catastrophic forgetting. New research by our team introduces Nested Learning (NL), a paradigm that views an ML model as a system of nested optimization problems. This approach unifies architecture and optimization, creating a deeper computational capacity for learning. This is a crucial step toward creating models with the continual learning abilities seen in the human brain. More in the blog by Vahab Mirrokni and Ali Behrouz: goo.gle/47IOCjP Read the NeurIPS 2025 paper: abehrouz.github.io/files/NL.…

all the questions and answers are contained within the understanding of "We're losing, people," which raises the questions: who is "we" and what is the exact perimeter of "people" , we speak about ? And just the implied phrase "in the process of" already allows us to exclude a large majority of the planet that has already lost a long time ago, well before we even started talking about generative AI :)

je vous ai decouvert a travers cette video reussie et je vais venir un peu voir ce qu'il se passe chez vous :) on attends les autres episodes now :)

it was a great one!

🐳 Part 3 is live! Learn how to modularize your @Docker Compose CI/CD pipeline for better organization & reusability. ✨ Use `extends` to break down jobs into separate files 📦 Improve maintainability & code clarity 📝 k33g.hashnode.dev/drasticall… #Docker #DevOps #CI

👇a superb paper 👇 Patterns for securing agents against prompt Injection arxiv.org/abs/2506.08837 🔖

:)

One nuance: because biotech outcomes are highly correlated (regulatory climates, volatility in drug approval regimes), you’d probably treat these bets as partially dependent. Read more 👉 lttr.ai/Aku02 #investing #methodology #frameworks

What GreyNoise Learned from Deploying MCP Honeypots greynoise.io/blog/deploying-…

I wrote a quick post with some thoughts on Android runtime instrumentation with Frida, looking at SQLite as a case-study. The main focus is on making instrumentation data more useful for scalable analysis. knifecoat.com/Posts/Runtime+…

raaah ... ca m'embete , en recuperant celui la ... je me suis apercu que j'en retrouve pas un qui etait pas dedans .... :/

I'm consulting for several companies that are using open-source models. There are 3 common problems they have to deal with: • Latency spikes • Costs are unpredictable • Compliance is a headache These are hard problems to solve, but there's light at the end of the tunnel: Token Factory just came out (launched by @nebiustf, one of my long-time collaborators). It looks like bullseye for companies that are dealing with AI at a production scale: • It provides dedicated inference • Sub-second latency and 99.9% uptime • Zero-retention policy (SOC 2 Type II, HIPAA, ISO 27001) • Supports fine-tuning and deployment without worrying about infrastructure. If you're building with open models, Token Factory appears to be the kind of platform that can help a ton. You can check it out here: tokenfactory.nebius.com/

Blog post: How to support FreeBSD on your cloud: daemonology.net/blog/2025-11…

Making solid progress. Any questions or concerns?