Technology & AI
AHA red-team agent finds reusable flaws in Claude Code, Codex
New arXiv preprint introduces AHA, a system using one AI agent to auto-discover reusable vulnerabilities in production agents like Claude Code and Codex.
Marcello Babbili
AI and technology, explained through what they actually mean. An AI-assisted newsroom under human editorial rules — every story cites its primary sources so you can check them yourself.
306 stories
Technology & AI
FactorDiff routes diffusion experts by pixel on ARC-AGI
FactorDiff decomposes diffusion samples into pixel-level factors and routes each to the best expert, beating global weighting on ARC-AGI reasoning tasks.
Technology & AI
AMT-X multi-turn red teaming hits 100% attack success on frontier LLMs
AMT-X multi-turn red-teaming framework hit up to 100% attack success on six frontier AI models, exposing critical gaps in current LLM safety testing.
Technology & AI
Playful AI email tone lifts positivity in 16,880-email field study
A field experiment across six companies found GPT-5's playful rewriting lifted emotional tone, with engagement gains flowing through an indirect path.
Technology & AI
PromptGraph turns LLM prompts into privacy graphs
PromptGraph models LLM prompts as graphs to catch privacy leaks that existing sanitisers miss, affecting anyone sending sensitive data to cloud AI.
Technology & AI
ARMOR-IMC preprint targets 50% accuracy loss in AI memory chips
ARMOR-IMC, an arXiv preprint, hardens in-memory computing chips against manufacturing faults and power attacks without retraining the AI model.
Technology & AI
EvoCUA-1.5 hits 63.2% on OSWorld with online RL training
EvoCUA-1.5 teaches AI agents to use computers through trial-and-error in sandbox environments, reaching 63.2% on OSWorld-Verified.
Technology & AI
Best AI agents fail 50% of visual tool tasks, Apple benchmark shows
A new open benchmark with 500+ tools reveals even frontier AI models can't reliably read an image and act on it, with most failures traced to seeing, not
Technology & AI
HiFi-LLP cuts neural architecture search time 8.6×
A new latency predictor knows when it's wrong, routing uncertain guesses to real hardware and slashing the cost of designing AI models for edge devices.
Technology & AI
AI vision models encode correct count but output wrong numbers
Vision-language models internally encode the right answer but misread it, a new arXiv study finds. A simple fix lifts counting accuracy 15.6 points.
Technology & AI
Spotify study links narrator voice to audiobook appeal
Spotify and Queen Mary researchers found voice features like tone and pace correlate with audiobook engagement even after controlling for the title.
Technology & AI
Quantum learning separation proved for many-body dynamics
Quantum learning separation proved in July 2026: a quantum procedure predicts many-body dynamics no classical algorithm can match, affecting quantum ML.