AI Safety news and analysis on Web Pulse: 10335 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A practitioner's field guide details five distinct prompt injection attack classes that evade simple keyword filters, including instruction smuggling through metadata, cross-document poisoning in RAG systems, indirect in…
In the race to artificial superintelligence (ASI), the first mover could outgrow rivals by colonizing space and using extraterrestrial resources, making international law irrelevant. Even a slight head start in robotic e…
A panel moderator in Toronto described jailbreaking an iPhone as 'rampant theft of IP,' sparking debate over digital sovereignty. The author argues that Canada's 2012 Copyright Modernization Act, which criminalizes jailb…
Seemplicity launched AI Analysts for exposure management and response, autonomous agents that replace manual vulnerability triage by conducting evidence-based exploitability investigations within remediation workflows. T…
Cloudflare released an open-source library of agent skills that enables AI agents to plan, deploy, manage, and migrate Zero Trust environments. The skills include migration logic for moving from Zscaler and Palo Alto Net…
The Washington Post tested six major AI chatbots on political questions and found systematic differences in ideological slant. ChatGPT and DeepSeek leaned left in most responses, while Gemini offered balanced answers 93%…
Accenture AI lead Stacy Pence said Australia is catching up in sovereign AI capabilities while leveraging offshore opportunities, as banking regulator APRA warned the nation is entering a dangerous period in AI adoption.…
Enterprises are entering a new era where AI agents, copilots, and autonomous systems actively perform work, raising critical questions about balancing automation, trust, and the human experience. IT leaders must evolve i…
A glitch in Google's Gemini AI exposed its internal reasoning process, revealing a raw thought stream that included translating Japanese subtitles for a Doraemon episode. The incident highlights transparency and safety c…
Vincent Schmalbach argues that as AI coding tools become more prevalent, traditional software testing—often considered tedious—grows in importance for ensuring code reliability and catching errors that AI might introduce…
Enterprises are moving workloads back to private and sovereign clouds to control costs and data, but the software supply chain underlying all cloud platforms is unprepared for AI-driven vulnerability discovery at scale. …
Stellar Cyber released updates 6.5 and 6.6, enhancing its AI-driven SOC platform with governed AI workflows, improved Auto Triage, sharper detections, and faster self-service data onboarding for MSSPs and enterprise secu…
A team at an unnamed company built a customer support agent on LangChain that hallucinated a wrong return policy in a multi-step process, logging success while being confidently wrong. This incident highlights the observ…
Equixly released its June 2026 product update, introducing AI Red Teaming for testing large language models, an Attack Surface Graph for visualizing exposure, Discovery upgrades including scheduled scans and MCP server d…
OpenAI quietly upgraded all free ChatGPT users to GPT-5.5 Instant, reducing hallucinations by 52.5% and improving multi-turn reasoning, now serving 900 million weekly users. The move narrows the gap between free and paid…
Agentic AI is reshaping financial crime compliance by enabling autonomous reasoning, investigation, and workflow execution, according to Flagright CTO Madhu Nadig. The shift moves beyond efficiency gains to fundamentally…
A survey of 600 UK HR professionals found that 91.6% regret redundancies made in the last 12 months, with 75% saying the cuts cost more than they saved. Many organizations rehired for eliminated roles within six months, …
Software developers face a transformed profession by 2026 as AI-generated code becomes the norm, shifting their role from writing code to specifying intentions and validating outputs. The rise of 'vibe coding' and AI too…
The Five Eyes intelligence alliance warned that frontier AI systems will transform cyber capabilities within months, lowering barriers for malicious actors while strengthening defenses. The joint statement from Australia…
A developer built TraceProof, an open-source Rust system that creates signed, verifiable records of AI agent decisions to address regulatory compliance challenges in banking. The system uses cryptographic signatures to e…