Researchers exploited GitHub’s AI workflows via prompt injection, leaking private repo data without authentication.
// curated from Hacker News with AI
Researchers exploited GitHub’s AI workflows via prompt injection, leaking private repo data without authentication.
Robostral Navigate enables robots to autonomously navigate complex environments using only a single RGB camera, outperforming multi-sensor models.
Microsoft's Flint simplifies AI data visualization with a semantic, adaptable language supporting 46 chart types across multiple backends.
SWE-1.7 reaches near GPT-5.5, leveraging advanced RL, multi-cluster training, data curation, and self-compaction for enhanced capabilities.
OpenAI's GPT-5.6 Sol, Terra, and Luna launch globally Thursday, expanding preview access for users worldwide.
Anthropic’s Fable misclassifies prompts related to bioinformatics, cybersecurity, and math, making it ineffective for research tasks.
Grok 4.5 outperforms on speed and cost; Claude Fable and Opus excel in reliability; GPT-5.5 is fastest but pricier.
GeoSQL enables Claude, Codex, and Copilot to perform advanced geospatial data analysis locally and self-hosted, with 4x task efficiency.
Databricks' internal benchmark shows open and proprietary models' performance on real coding tasks, highlighting cost, efficiency, and best practices.
Fortress is an open-source, C++ patch-based Chromium engine that tricks anti-bot detection by mimicking native browser signals internally.
The AI boom mirrors 2008's credit cycle, driven by decelerating growth and hidden risks in debt structures, not just level or velocity.
Kastor offers Terraform-style, declarative specs for defining, validating, and planning AI agents with code generation, not runtime execution.
The AI industry is a fragile bubble of hype, overvaluation, and circular financing that should be allowed to collapse without bailouts.
Onboard-CLI uses AST parsing and React Flow visualization to map, analyze, and enforce architecture boundaries in large codebases.
AI-heavy workflows reveal testing flaws, highlight variance issues, and showcase agentic coding's efficiency despite poor LLM test quality.
Ghostmeet is a self-hosted, private AI meeting tool that transcribes, summarizes, and runs entirely locally, ensuring data privacy.
Foreman is a self-hosted, cost-aware, secure LLM gateway that routes requests based on policies, ensuring cost efficiency and model quality.
ZML/LLMD is a cross-platform LLM inference server supporting popular models and modern features with native cloud storage support.
Abraló simplifies managing multiple Claude Code agents, fixing login issues, offering a lightweight, fast, cross-platform UI.
Most nurses distrust AI for patient care, citing inaccuracy; only 11% believe it will help with staff shortages soon.
ChatGPT-generated flyers flood communities, blending generic designs with bold text, causing frustration among creatives and small businesses.
AI is now controlling the stock market, with the bond market likely to follow, signaling widespread automation in finance.
DualView defends personal AI agents like OpenClaw from indirect prompt injection by isolating and controlling untrusted data.
Monogram: AI app with interactive visual interface for personalized recommendations and content generation.
Brown professor suspects widespread AI cheating, cancels final, reveals low scores, and criticizes university response.
Moo creates hardware-isolated Linux machines per git branch, snapshots states, aiding parallel coding, debugging, and time travel.
Anthropic extends free access to Claude Fable 5 until July 12; after that, usage shifts to costly prepaid credits.
AI design quality improves through self-review, example comparison, and targeted feedback; measurement remains subjective.
User blocked by security; file a ticket for review.
AI agents face new security risks from Data Injection Attacks that bypass defenses, causing unwanted actions and vulnerabilities.
FTC aims to prevent AI companies from misleading consumers by steering outputs away from stated objectives, ensuring honest representations.
The Trump administration eases restrictions on OpenAI's GPT 5.6, potentially boosting AI development and deployment.
FactIQ offers real-time finance and economic data via a plugin, enabling AI agents to analyze markets using official statistics.
Tarit is a fast, self-hosted hypervisor and sandbox cloud for AI agents, enabling isolated, ephemeral microVMs with easy orchestration.
Trump accelerates approval of gas plants to power AI data centers, overriding environmental concerns and banning wind energy.
AI agent skill generates accurate UML class diagrams from descriptions or code, embedding validated SVG images into docs.
Dex is an AI toolkit that automates schema evolution, data exploration, transformation, and cost management for analytics engineering.
AI-generated content is often dismissed as low-quality "slop," prompting humans to focus on storytelling, imperfection, and authenticity to stand out.
China uses AI for covert propaganda, censorship, and influence operations, skewing global narratives in Beijing’s favor.
IBM and Red Hat launch Lightwell to automate and secure open-source code against AI-driven vulnerabilities.
China warns developers to remove Claude Code over fears of backdoor data collection and unauthorized user monitoring.
Enables DuckDB to connect and query any database with an ADBC driver, using Apache Arrow for fast data transfer.
OpenAI's Deployment Company acquires Northslope to boost enterprise AI deployment with hundreds of FDEs for critical operations worldwide.