Tag: AI safety

A new AI & Society article argues that the failures of autonomous LLM agents documented in live deployments stem from ...

At the 13th Heidelberg Laureate Forum, 21 laureates and 200 young researchers debated AI's accelerating role in mathematics, quantum-era cryptographic ...

A new review in Artificial Intelligence Review finds that global AI assurance frameworks remain fragmented and under-specified, proposing five cross-cutting ...

Researchers have demonstrated a switchable backdoor attack that injects dual tokens through the layers of Vision Transformers, achieving up to ...

A new theoretical review argues that reward hacking, benchmark failure, and distribution shift share a single structural root: observation regimes ...

Researchers argue that multi-step agentic AI systems break traditional clinical validation models and propose structured testing harnesses adapted from simulation, ...

A systematic review of 93 studies reveals eight attack types and nine defence types across six categories of unsupervised machine ...

Researchers have developed a hypernetwork-based framework that lets continually learning AI systems erase specific tasks without access to the original ...

A new Nature Machine Intelligence study provides causal evidence that language models actively use internal confidence signals to shape their ...