AI Safety news and analysis on Web Pulse: 13852 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
On May 25, 2026, Pope Leo XIV published "Magnifica Humanitas," a 235-page encyclical addressing AI's impact on human dignity, labor, warfare, and concentrated power, co-presented with Anthropic co-founder Christopher Ola…
Researchers bypassed safety guardrails on Meta and Google's AI models within minutes, exposing vulnerabilities that allow the generation of harmful content. The findings raise concerns about the effectiveness of current …
Confluent Cloud introduces new capabilities for building secure, compliant Retrieval-Augmented Generation (RAG) architectures in regulated industries and the public sector, addressing data residency, PII leakage, halluci…
A model build failed to explain why an opt-out record appeared in its training data, despite the dataset hash being correct. The incident revealed that the team's manifest lacked source policy, exclusion reports, and rev…
Armos has released a local PII detection and masking layer for LLM APIs that runs entirely on the user's server, with no data sent to third parties during detection. The open-source library uses Microsoft Presidio and sp…
Pope Leo XIV issued his first encyclical Monday, calling for robust regulation of artificial intelligence and denouncing the technology as an "instrument of domination, exclusion and death" when driven by profit and powe…
India's legal framework, built on statutes enacted before the AI era, is struggling to keep pace with technologies that can generate art, clone voices, and create convincing deepfakes. With no dedicated AI or deepfake le…
David Ohnstad outlined strategies for product teams to build user trust through transparency, ethical data practices, and privacy protections in AI-driven development. The approach emphasizes clear communication about da…
Microsoft has introduced MDASH, a multi-model agentic security platform that uses over 100 specialized AI agents to automate large-scale vulnerability discovery across Windows, Hyper-V, and Azure codebases. The system ac…
A new analysis of artificial intelligence's potential trajectories identifies three distinct future scenarios—long timelines, optimistic short timelines, and pessimistic short timelines—each requiring different strategic…
Noumenon AI released AutoMaxFix, an open-source Python CLI tool that creates a controlled, safety-gated repair loop for AI-generated software. The tool converts failing tests and bug reports into structured tickets, then…
AgentGate, a new trust authorization layer for autonomous AI agents, intercepts every agent action before execution to verify identity, validate delegation chains, and detect behavioral drift in real time. The system blo…
AMAS (AI Multi-Agent Security Analysis System) uses causal reasoning to teach AI models to understand logical vulnerabilities like IDOR, privilege escalation, and workflow bypasses, rather than relying on pattern matchin…
The "Stepford Wives" narrative of men replacing women with compliant, perfect versions has evolved into a modern technological horror: AI girl Friday chatbots designed to organize lives, offer constant praise, and never …
A developer has outlined a minimum viable test suite for teams working with AI coding agents, arguing that strategic coverage at three key "seams"—API contracts, data layer boundaries, and critical business rules—provide…
Internet users used an AI coding assistant to reconstruct audio of three dead pilots' final moments from a spectrogram in a federal crash investigation, violating a 1990 law that prohibits releasing cockpit voice recordi…
Pope Leo XIV released his first major papal encyclical, *Magnifica Humanitas*, on Monday, warning of the risks of artificial intelligence and calling for new legal and ethical frameworks to govern the technology. The doc…
Credential brokering is an emerging security paradigm that allows AI agents to use credentials for external systems without directly accessing them, addressing the risk of credential exfiltration through prompt injection…
Apple blocked $11.2 billion in potentially fraudulent App Store transactions over the past six years, including $2.2 billion in 2025 alone, while also rejecting over 2 million app submissions and removing nearly 200 mill…
Mimesis, an open-source data generation library, can be used to create balanced, counterfactual datasets for auditing bias in machine learning models. A hands-on guide demonstrates training a loan approval classifier on …