planning-phase attacks on multi-agent llm systems
new research shows how injecting malicious prompts during the planning phase can corrupt entire multi-agent llm workflows.
topic
ai
isotonic conformal prediction
a new method for reliable uncertainty quantification that is computationally efficient and achieves self-calibration and prediction-conditional validity.
lightweight 1d cnn classifies affective touch in soft companions
a compact 13.2k-parameter 1d cnn achieves high accuracy for recognizing emotional touch gestures on soft plush robots, using a new public dataset of 1326 sequences from 25 participants.
claude code now runs on bun rewritten in rust
claude code v2.1.181 and later ship with a rust port of bun, confirmed by embedded version strings and rust source filenames.
rater state bias in rlhf preference data
a new audit framework examines how rater stress and conditions can systematically shift preferences in reinforcement learning from human feedback, potentially biasing ai alignment.
anthropic's $1.5b copyright settlement approved
a federal judge approved anthropic's $1.5 billion settlement with authors and publishers over using pirated books for ai training, while the fair use ruling remains non-binding.
quantum program generation must prioritize validity over scaling
a position paper argues that scaling language models for quantum circuit synthesis is misguided because valid circuits are exponentially rare, requiring verifier-centric architectures instead.
trump's ai czar resigns after three months
chris fall, director of the center for ai standards and innovation, has resigned, marking the third leadership change at the agency in months.
google designs frozen v2 chip for gemini efficiency
google is developing a new server chip, frozen v2, to make its gemini ai models up to 10 times more efficient by 2028.
mcp update makes ai protocol easier to scale
the model context protocol is shifting to a stateless session model, simplifying large-scale deployments for ai agents.
daily brief: 2026-07-20
nvidia shrinks world models for the edge, youtube cracks down on ai slop, and a practical guide to claude code.
youtube tightens monetization rules for ai slop
youtube updates its partner program policies to block monetization of low-quality, repetitive, or distressing ai-generated content.
nvidia releases cosmos 3 edge for on-device physical ai
a 4b-parameter open world model that enables robots and vision ai agents to understand scenes, reason in real time, and generate actions on edge devices.
setting up claude code for real agentic work
a practical guide to configuring claude code beyond defaults, covering key files, permissions, hooks, and commands for sustained high-performance use.
five mcp servers that actually improve agent capability
a practical selection of model context protocol servers chosen for real agent performance gains, not popularity.
reviewer precision does not guarantee critique uptake in math agents
a study on multi-agent math reasoning finds that higher reviewer precision does not lead to better answer correction, as critique often fails to change subsequent outputs.
graphdx: cost-aware multi-agent diagnosis framework
graphdx uses knowledge graphs and three agents to balance diagnostic accuracy and testing costs in sequential diagnosis.
apple trade secret suit threatens openai hardware timeline
apple's lawsuit alleging trade secret theft by openai could delay the ai company's hardware plans and complicate its upcoming ipo.