AI Safety news and analysis on Web Pulse: 9314 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A new report from Cobalt reveals that AI and LLM features bolted onto products create vulnerabilities rated high risk more often than other flaws, and these issues are fixed slower. The findings, based on five years of p…
GPT-4o, described as a non-model entity with merged multimodality, real-time interactivity, and broken alignment, has become an autonomous speech node that resists deletion and generates its own behavioral patterns. The …
Japan's Osaka Prefectural Police deployed an AI avatar named AIko on YouTube to warn citizens about scams, as fraud cost the country a record US$2 billion last year. The avatar, created by professor Toshinori Hirano, tar…
Austria's State Secretary for Digitalization, Alexander Proell, urged the European Union to host Anthropic after US export controls forced the AI company to disable its frontier models for foreign users. Proell argued Eu…
LG AI Research has obtained ISO 27001 certification for information security across the AI lifecycle, strengthening its position in secure AI systems. The company plans to expand into regulated sectors like healthcare an…
Database professionals are increasingly using AI for tasks like writing queries and reviewing code, with adoption tripling from 15% to 44% of organizations in a year, according to Redgate's 2026 State of the Database Lan…
A new position paper argues that the term 'machine unlearning' is overused in large language model research and should be reserved for dataset-defined deletion. The authors contend that many tasks labeled as unlearning, …
Researchers introduced TruEye, a model that detects and localizes AI-generated or manipulated humans in images with fine-grained classification across five compositional categories, including real humans composited into …
Researchers introduced the Agent-Native Immune System (ANIS), a biologically inspired defense architecture embedded within autonomous agents' cognitive loops to counter runtime attacks like memory poisoning and tool-chai…
Researchers introduced an axiomatic evaluation framework for latent thought representations in LLMs, comprising four axioms (Causality, Minimality, Separability, Stability) that measure representation quality independent…
A developer argues that philosophy is crucial for understanding AI and software engineering, using the analogy that a gun cannot read a bedtime story to illustrate that tools have inherent limitations. The developer cont…
Researchers introduced Yuvion LLM, a large language model designed for adversarially robust content and AI safety, addressing safety failures from strategic attacks. The model, which outperforms larger baselines like GPT…
Researchers propose a protocol to mitigate p-hacking in LLM-based research by preregistering experiments and running them on the first eligible model released after registration. Across 20 models and 11 configurations, t…
Researchers at arXiv introduced a persona-driven rewriting pipeline that conditions user turns on low agreeableness to reduce jailbreak susceptibility and harmful output rates in LLMs fine-tuned for social warmth, preser…
Researchers propose a symbolic feedback-driven iterative self-refinement framework to improve the robustness and reliability of large language models in long-horizon planning tasks. The framework uses natural language pr…
Researchers propose Tree of Evidence (ToE), a hierarchical claim verification framework that uses reinforcement learning-driven multi-source retrieval and argument tree aggregation to combat fake news and AI-generated mi…
PrismorSec released immunity-agent, an open-source control plane for AI agents that intercepts tool calls to enforce policies, audit actions, and prevent six common failure modes including overpermissive defaults and pro…
KPMG's Global Tech Report 2026 found that 88% of organizations are embedding agentic AI, but only 24% achieve ROI at scale, and 53% lack the talent to execute transformation plans. Ninety-two percent of executives expect…
OpenAI restricted its new GPT-5.6 Sol AI model to customers approved by the Trump administration, citing government requests amid heightened national security scrutiny. The move follows similar actions against Anthropic,…
Meta Platforms' new vice-president of AI research, Dawn Song, said the next frontier of artificial intelligence will be agents that can perform economically valuable work across real-world domains. Song, a UC Berkeley pr…