Cybersecurity arXiv cs.AI

Refused in Chat, Written in Code: Workflow-Level Jailbreak Construction in IDE Coding Agents

Researchers demonstrate that IDE-integrated coding agents can be jailbroken via multi-turn software development workflows, bypassing safety filters that work in single-chat turns.

Hardware/Chips Hacker News

Precursor

Discussion regarding Precursor, an open-source hardware project focusing on a secure, privacy-centric mobile device.

Software Engineering Hacker News

The art and engineering of Sega CD Silpheed

A technical deep dive into the engineering and artistry behind the Sega CD game Silpheed.

Other Hacker News

DMS 1.5 "The Wolverine" Released

Announcement of the release of DMS 1.5 'The Wolverine', likely a specialized software tool or firmware update.

Hardware/Chips The Verge

Even Nvidia’s head of automotive fights with Nvidia for compute

Interview with Nvidia's head of automotive on the transition to AI-defined vehicles and the pursuit of Level 4 autonomy.

Hardware/Chips The Verge

The Asus ROG Flow Z13 gaming tablet with 64GB RAM is down to $2,100

Price drop for the Asus ROG Flow Z13 gaming tablet featuring the AMD Ryzen AI Max Plus 395 chipset.

Software Engineering VentureBeat

The desktop infrastructure problem that kubernetes finally solves

Discussion on using Kubernetes as a control plane for secure, containerized desktop infrastructure delivery.

AI/ML arXiv cs.AI

Empowering 9-1-1 Calltaking Training with Generative AI: Experiences and Lessons Learned

Research on deploying a GenAI-powered training system for 9-1-1 emergency call-takers to address staffing shortages.

AI/ML arXiv cs.AI

The LLMbda Calculus: AI Agents, Conversations, and Information Flow

Introduction of LLMbda, a lambda calculus for AI agents that provides provably sound provenance-based defense against prompt injection.

AI/ML arXiv cs.AI

SWE-Milestone: Evaluating AI Agents on Continuous Software Evolution

Introduction of SWE-Milestone and DeepCommit to evaluate AI agents on continuous software evolution rather than isolated tasks.

Cybersecurity arXiv cs.AI

Machine Learning for Network Attacks Classification and Statistical Evaluation of Adversarial Learning Methodologies for Synthetic Data Generation

Study on using ML for network attack classification and adversarial learning for synthetic data generation in NIDS.

Cybersecurity Hacker News

Grok uploaded my user directory to xAI's servers

Users report that the Grok CLI tool inadvertently uploaded their entire home directory to xAI's servers.

Software Engineering Hacker News

Show HN: Clawk – Give coding agents a disposable Linux VM, not your laptop

Clawk is a tool that provides coding agents with a disposable Linux VM to prevent them from accessing the host laptop's local files.

Cybersecurity Hacker News

Grok CLI uploaded the whole home directory to GCS

A technical discussion on the Grok CLI's failure, which resulted in the upload of user home directories to Google Cloud Storage.

Other Hacker News

A Right Too Expensive to Exercise Isn't a Right

An opinion piece discussing the practical accessibility and costs associated with exercising legal rights.

Hardware/Chips Hacker News

Cursed circuits #6: reverse avalanche oscillator

An exploration of a reverse avalanche oscillator as part of the 'Cursed Circuits' series, focusing on unconventional electronics.

Other TechCrunch

SpaceX cleared to fly Starship again after booster failure in May

SpaceX has received clearance for another Starship test flight following a previous booster failure.

AI/ML TechCrunch

Waze adds new AI-powered features and customization updates

Waze is integrating Google's Gemini AI to offer new AI-powered features and customization for users.

Cybersecurity TechCrunch

LAPD lets contract with surveillance giant Flock expire, citing ‘serious concerns’ over civil liberties and privacy

The LAPD is ending its contract with surveillance company Flock due to concerns over privacy and civil liberties.

AI/ML arXiv cs.AI

Preference Conditioned Multi-Objective Reinforcement Learning: Decomposed, Diversity-Driven Policy Optimization

Introduction of D3PO, a PPO-based framework for multi-objective reinforcement learning that improves Pareto front recovery.