PrivateGPT relaunches as a full production API, allowing developers to build private, local AI apps with RAG, MCP, and text-to-SQL capabilities — no cloud dependency required — backed by over 57,500 GitHub stars since going viral in 2023.
AI agents are vulnerable to a sophisticated two-stage prompt injection attack that exploits trusted internal tools like Jira, GitHub, and Salesforce, slipping past existing OWASP defenses by hiding malicious instructions in free-text fields and triggering unauthorized actions that no single security checkpoint can detect alone.
Google and Cathay Pacific are deploying AI technology to reduce aviation's climate impact, with early trials across 80+ flights achieving a roughly 40% reduction in contrail warming effects by guiding pilots to make slight altitude adjustments that avoid contrail-forming atmospheric zones.
ByteDance, with founder Zhang Yiming personally at the helm, is set to launch a real-time spatial video AI model for Pico headsets as soon as next month, rendering interactive 3D cloud environments at 50ms latency while outspending rivals three to four times as part of a massive AI expansion backed …
Two MMLU benchmark scores from the same model family are deemed incomparable despite their similar numbers, exposing a critical flaw in AI evaluation: sharing a benchmark name means nothing if the testing conditions differ, as variables like prompt format, grader type, and dataset splits can shift accuracy by several percentage …
Qwen-Drive-1.0 launches as a powerful unified AI model combining 3D perception, visual question answering, and motion planning for autonomous driving, outperforming competitors on major benchmarks and now freely available under an open-source license.
RAG systems face serious security threats as researchers reveal that just five crafted documents can corrupt AI responses with over 90% success, while embedding vectors previously considered opaque can now be used to reconstruct source text, forcing organizations to reclassify vector store breaches as significant data leaks.
A minimalist AI agent called hip-agent, built with just two tools and a self-readable core loop, outperforms OpenAI's Codex CLI on a 113-task coding benchmark, resolving 64.6% of DeepSWE challenges versus Codex CLI's 63.7%, proving that radical transparency in agent design can rival proprietary systems.
SuperAstra is a groundbreaking open-source tool that lets players modify SNES games in real time using natural-language AI prompts, integrating OpenAI's gpt-6-astra model with the BizHawk emulator to read and rewrite live game memory on the fly.
Herdr 0.9.0 launches as a free, open-source Rust-based terminal multiplexer purpose-built to keep AI coding agents like Claude Code and Codex running persistently in the background, even after SSH disconnections, with real-time status tracking and multi-agent coordination via CLI and socket API.