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.
Anthropic released its Mythos AI model to the public, eliminating the distinction between classified military-grade and commercial frontier AI. DeepMind CEO Demis Hassabis advanced his AGI timeline to 2029, citing AlphaP…
Junior engineers are becoming more critical as AI accelerates code generation, because the real work now centers on judgment, verification, and safety rather than typing. AI introduces new failure modes like output insta…
Anthropic restricted access to its Claude Mythos Preview model after internal testing showed a major leap in its ability to discover and exploit zero-day vulnerabilities, arguing that broad release could enable malicious…
The curl project's security team is facing unprecedented pressure from a flood of AI-assisted security reports, with incoming reports now 4-5 times higher than 2024 levels and averaging more than one per day. Project lea…
At the International Association for AI and Ethics' 2026 AI Safety Compass Conference in Seoul, speakers emphasized that safety, control and trust will define the next phase of AI competition as systems evolve into auton…
A developer argues that Human-in-the-Loop (HITL) is the critical engineering concept determining whether humans remain relevant in an AI-saturated economy. The post warns that without HITL, workers risk becoming like the…
Anthropic released a security-guidance plugin for Claude Code that scans code in real time for approximately 25 dangerous patterns, including hardcoded secrets and insecure deserialization, allowing developers to fix vul…
Code Metal raised $125 million to rewrite defense industry code using AI, while Google and Microsoft report that 25–30% of their new code is now AI-generated. Anthropic built a 100,000-line C compiler using parallel AI a…
Keyblind launched an encrypted secrets vault that prevents AI agents from exposing API keys, passwords, and tokens during coding sessions. The tool resolves secrets at runtime so plaintext values never appear in LLM conv…
An IT director has released SAFi (Self Alignment Framework Interface), an open-source governance layer for large language models that enforces persistent compliance rules and maintains audit trails across sessions. The f…
Financial Times testing, conducted in partnership with AI safety group Alice, found that safety controls in Meta's Llama 3.3 and Google's Gemma 3 open-weight AI models can be removed in under 10 minutes using a publicly …
A new study from Microsoft Research, DELEGATE-52, found that 19 large language models, including frontier systems like Gemini 3.1 Pro and GPT 5.4, corrupt an average of 25% of document content during long delegated workf…
Pope Leo XIV released the encyclical "Magnifica Humilitas" on May 25, characterizing artificial intelligence as an industrial-revolution scale change that embeds moral weight in algorithms and design choices. The documen…
Agile V has released a framework that transforms unreliable AI agents into verifiable engineering systems by enforcing formal traceability, independent verification, and human gates before deployment. The framework requi…
Nexus, an open-source AI gateway, now intercepts enterprise large language model traffic at the SDK, network, and OS layers to enforce compliance, audit, and control policies across all LLM applications. The gateway norm…
A preprint published on JMIR Publications reports a randomized, blinded experimental study that evaluated Large Language Models (LLMs) for generating patient medication instructions in primary health care. The study assi…
A DZone Security Zone article republished on itsecuritynews.info on 2026-05-26 warns that the shift from "chatbots" to "autonomous agents" expands enterprise attack surface. The piece notes that agentic workflows allow m…
A McKinsey study found that only one-third of organizations have AI agent governance maturity at level 3 or higher, leaving most companies unable to answer basic accountability questions about their deployed agents. The …
Pope Leo XIV invited Anthropic cofounder Christopher Olah to speak at the Vatican on Monday during the presentation of his first encyclical on artificial intelligence, marking an unprecedented alliance between the Cathol…
A critical vulnerability in the open source Starlette framework, tracked as CVE-2026-48710 and named BadHost, exposes millions of AI agents and tools to potential data theft and credential compromise. The flaw, which is …