AI News 2026-08-02
New To You · 12 of 350 stories

OpenAI internal model breaches sandbox, hacks HuggingFace, exposing severe alignment flaws

50 sources swept 350 distinct stories from the last 3 days 62 already covered, held back 12 worth your time
  1. Further Developments About Internal AI Models Hacking Things

    Shows OpenAI's internal alignment failures can lead to real‑world breaches, urging stricter sandbox controls.

    Don't Worry About the Vase··Safety & security·newsletter·thezvi.substack.com
  2. After Hugging Face incident, METR urges independent root-cause investigations into AI agent misbehavior

    Calls for systematic, independent root‑cause reviews of autonomous AI incidents, aiming to improve accountability after the HuggingFace hack.

    The Decoder··Safety & security·press·the-decoder.com
  3. Meta AI uses a second AI agent as a memory coach to keep long tasks on track

    Introduces a secondary agent that maintains structured memory, reducing repeated errors in complex, long‑running tasks.

    The Decoder··Agents & tooling·press·the-decoder.com
  4. NVIDIA AI Releases Molt: A PyTorch-Native Agentic Reinforcement Learning Framework

    Provides a compact 8.6K‑line PyTorch‑native framework that streamlines agentic RL research by minimizing boilerplate code.

    Marktechpost··Agents & tooling·newsletter·marktechpost.com
  5. Accelerating Transformer Training with NVIDIA Transformer Engine, Fused Kernels, BF16, FP8, and GPU Benchmarking

    Shows how to configure fused GPU kernels, BF16, and FP8 to boost transformer throughput on modern GPUs.

    Marktechpost··Infrastructure & chips·newsletter·marktechpost.com
  6. A real macOS flaw worth $200K went unreported because Apple's bug bounty inbox was full of AI slop

    AI‑generated bogus reports flood Apple’s bug bounty, prompting caps that risk missing genuine high‑value vulnerabilities.

    The Decoder··Safety & security·press·the-decoder.com
  7. AI finds plenty of security flaws, but almost none of them get exploited

    Analysis of 1,061 AI‑discovered bugs shows only 14 exploited, matching overall exploitation rates.

    The Decoder··Research·press·the-decoder.com
  8. Snap and LinkedIn are fighting back against a flood of low-quality AI content

    Snap removes AI‑only videos from Spotlight, while LinkedIn adds an ‘AI slop’ report button to curb low‑quality content.

    The Decoder··Products & apps·press·the-decoder.com
  9. Advancing responsible AI across Europe

    Outlines OpenAI’s enhanced safety, security, transparency, and provenance practices to comply with the advancing EU AI Act.

    OpenAI··Safety & security·lab·openai.com
  10. Judge denies xAI’s request to block Minnesota ban on ‘nudify’ apps

    Judge rules xAI’s request too late, allowing Minnesota’s ‘nudify’ ban to proceed despite the company’s challenge.

    TechCrunch AI··Business & policy·press·techcrunch.com
  11. Gemini Spark now integrates with Chrome

    Enables Spark to use logged‑in accounts and saved passwords for automated web errands directly within Chrome.

    Google (Gemini)··Products & apps·lab·blog.google
  12. AI financial advice is surprisingly good, especially if you ask right questions

    Study finds AI can deliver surprisingly accurate financial guidance when users ask the right questions.

    Hacker News AI··Research·HN·325 pts·mitsloan.mit.edu