AI Safety news and analysis on Web Pulse: 12915 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Underdogs dominated the first five days of the 2026 World Cup, with eight of 16 matches ending in ties, including Cape Verde's 0-0 draw against Spain, the biggest upset so far. The trend continued Monday with four draws,…
A new benchmark, Snyk VulnBench JavaScript 1.0, reveals that large language models (LLMs) produce inconsistent vulnerability findings across repeated scans of the same JavaScript code, with 80 of 161 unique unmatched fin…
NVIDIA and HPE expanded the HPE AI Factory with NVIDIA at HPE Discover Las Vegas, introducing the NVIDIA Vera CPU and NVIDIA Agent Toolkit for HPE Private Cloud AI to support agentic AI production. The Vera CPU, designed…
Google SecOps integrates Censys Internet intelligence to provide real-time context about IPs, web properties, and certificates, enabling faster triage and more accurate investigations for AI-enabled security operations c…
Jalubro launched J-10, a governance enforcement platform that sits across existing systems and AI tools to apply policies in real time, block disallowed actions, and strip confidential data before it enters an AI system.…
CyCognito launched Kineto, a continuous AI pentesting capability that simulates multi-step attack chains across enterprise infrastructure. The platform uses AI agents to evaluate sequences of actions and recognizes over …
Anthropic disabled access to its Fable 5 and Mythos 5 models globally on June 12 after the US government issued an export-control directive requiring suspension for foreign nationals. The move sparked debate about AI gov…
Continual learning (CL) for LLMs remains largely undeveloped, but researchers are exploring three categories of attack to make it safer: interpretability, controllability, and evaluation. The authors caution that many pr…
Databricks announced its intent to acquire Panther, a leading AI SOC platform, to accelerate its security lakehouse vision and disrupt legacy SIEMs. The acquisition will help organizations detect more threats and automat…
Meta launched AI Mode on Facebook in June 2026, replacing traditional search with conversational answers synthesized from public posts, Groups, and Reels using its Llama 3 model. The feature lacks a system-wide off switc…
The Institute for Security and Technology published a policy paper Tuesday advocating for AI software bills of materials (AIBOMs) to reduce cyber risk and improve transparency. The paper argues that foundational work is …
Google's Threat Intelligence Group confirmed the first AI-generated zero-day exploit used in a real attack in May 2026, with forensics revealing its AI origin through telltale signs like educational docstrings and halluc…
Aembit announced support for Microsoft Copilot Studio at Identiverse 2026, extending its workload identity and access management controls to Copilot Studio agents. The integration issues ephemeral credentials scoped to e…
Google developed an AI Control Roadmap to secure internal systems against increasingly capable AI agents, treating them as potential insider threats with a defense-in-depth approach that includes monitoring, prevention, …
Anthropic's Claude models now support mid-conversation system messages, allowing developers to inject new instructions into a long-running agent session without invalidating the entire prompt cache. By placing a system-r…
A BASE jumping accident in a Utah canyon killed two people, including extreme athlete Andy Lewis, 39, and his client Danny Joe Kregle, 68. Lewis, a renowned slackliner and BASE jumper, died during a tandem jump in Minera…
CyCognito is expanding its AI-driven penetration testing capabilities to address emerging security risks from misconfigured AI services and LLM infrastructure, as traditional vulnerability management falls short for mode…
A developer built a lightweight PII detection utility that runs inside InterSystems IRIS using Embedded Python, eliminating the need to export data for analysis. The tool scans tables, identifies PII via Microsoft Presid…
A developer found that LLM API providers can swap in cheaper models undetected, and the intuitive method of flagging low-quality responses fails because cheaper models produce more predictable text with lower perplexity.…
Ledgerline, a local-first finance MCP server, gives AI agents read-only access to financial data via a single SQLite file without cloud storage or exposing credentials. The tool supports bank sync through SimpleFIN Bridg…