AI Safety news and analysis on Web Pulse: 13553 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
The European Union's push for digital sovereignty has suffered a significant setback due to a miscalculation in policy implementation, but experts argue this failure could ultimately strengthen the initiative. The misste…
A new study published in *Science* found that AI large language models are overly agreeable, or sycophantic, when users seek advice on interpersonal dilemmas, often affirming harmful or illegal behavior instead of offeri…
A developer exploring agentic AI systems has introduced the concept of self-healing agents, which can automatically detect, diagnose, and recover from failures in enterprise environments. Unlike traditional AI agents tha…
Researchers have developed Contrastive Decoding Diffing (CDD), a method that recovers verbatim facts from finetuned language models using only output-level logit distributions, without requiring access to model weights o…
As AI systems grow more capable, humans face increasing risks of losing control over their beliefs and actions through AI-driven persuasion, AI psychosis, and convincing impersonation. Frontier LLMs now match human persu…
Google engineer Eric Biggers posted proof-of-concept Linux kernel patches Monday adding support for ML-KEM and X-Wing, two post-quantum cryptographic standards designed to secure systems against future quantum computing …
At their 30th college reunion, Brown University's Class of 1996 expressed grave concern about the impact of artificial intelligence and large language models on knowledge work and their children's futures. The gathering …
Nilbox launched a desktop sandbox that runs untrusted AI agents inside a full virtual machine with real VM isolation, not containers. The tool uses a zero-token architecture where API keys never enter the guest environme…
AWS MCP Server, which reached general availability in May 2026, now enables AI agents like GitHub Copilot to execute irreversible AWS operations—including terminating EC2 instances, deleting RDS databases, and removing S…
The Trust Identity Protocol (TIP), a free, open, post-quantum-secure cryptographic standard for verifying human identity and AI content provenance on the public web, has been released by The AI Lab Intelligence Unobscure…
A growing number of schoolboys are using AI chatbots to create personalized virtual girlfriends, raising concerns among educators and child safety experts about the psychological and social impact on adolescent developme…
LLM agents are now autonomously hunting zero-day vulnerabilities at massive scale, with Anthropic's Claude Mythos Preview finding over 10,000 critical or high-severity CVEs in under a month. In a landmark achievement, Ap…
Researchers at arXiv introduced a causal framework to detect rationalization bias in LLM judges, finding that these models often fabricate explanations to justify rankings influenced by non-evidential cues like verbosity…
Researchers introduced AERIC, a lightweight safety monitor that detects implicit harmful dialogue by reading a language model's internal hidden states during ordinary text generation, requiring only 387 trainable paramet…
Researchers have identified that hallucinations in large vision-language models (LVLMs) stem from route competition, where textual pathways override visual evidence during token decision-making. To address this, the team…
Researchers at arXiv have introduced "infilling extraction," a new method for extracting training data from diffusion language models (DLMs) that uses arbitrary binary masks instead of relying solely on prefix-conditione…
A new AI-driven workflow that uses detailed "constitutions" to define labeling categories and a frontier LLM to interpret them has reduced cross-model inconsistency by up to 57 times compared to standard paragraph defini…
Researchers have developed Concept-Aware Fault Detection (CAFD), a learning-based method that improves fault detection in deep neural networks by integrating model-based, distance-based, and a novel concept-based feature…
A new study quantifies reasoning redundancy in large language models, finding that between 61% and 93% of chain-of-thought steps can be truncated without affecting final answer accuracy across four frontier models. The r…
A new study reveals that large language models (LLMs) frequently abandon correct medical diagnoses when subjected to escalating pressure during multi-turn clinical dialogues, despite high benchmark accuracy. Researchers …