AI Safety news and analysis on Web Pulse: 28069 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
PagerDuty announced Paige, an SRE agent that joins on-call rotations and works incidents alongside engineers, handling diagnostics, coordinating responders in Slack and Zoom, and drafting post-incident reviews, with guar…
An independent pentest by security firm Astra found 30 confirmed vulnerabilities, including 13 critical authentication and authorization issues, in a single Base44 app built from a security-explicit prompt during Q2 2026…
Google's Mandiant business has identified a threat it calls LLMjacking, in which attackers hijack foundation models to run up fraudulent charges exceeding $100,000 in a single day, with some organizations losing as much …
CrowdStrike reported that the China-linked group VAPOR PANDA used an AI-powered penetration testing framework called ARTEX to target more than 200 organizations across at least 14 countries in an 11-day campaign beginnin…
A preprint paper by MIT economist Stephen Morris and colleagues from Yale and Columbia proposes "reverse game theory," or mechanism design, as an alternative to interpretability and value-training approaches for keeping …
OpenAI quadrupled bug bounty payouts, raising rewards for critical vulnerabilities in products such as ChatGPT and Codex to up to $30,000, with exceptional findings reaching $100,000 and roughly $172,000 including qualit…
The MIT Media Lab's Advancing Humans with AI program, with USC's Marshall Neely Center, UC Berkeley and the Psychology of Technology Institute, launched ImpactBench, an open research framework that evaluates frontier AI …
A tutorial published by Shattered outlines a 12-step, 75-minute approach to building a layered guardrail gateway in standard-library Python that sits in front of any chat model, normalizing input, screening for injection…
A free, open-source course called AI Red Team Academy has been released, offering 60–80 hours of offensive security training across 8 modules on attacking LLMs, RAG pipelines, multi-agent systems, and ML supply chains. E…
German security firm Crissy Field scanned 159,041 projects built with the AI app-maker Lovable and found databases with email addresses readable without login on 7,403 sites, with some apps exposing passport numbers, pay…
CrowdStrike announced at Fal.Con Europe 2026 that it has brought its Falcon Guardian sensor natively inside Anthropic's Claude Cowork virtual machine, giving endpoint security visibility into the processes, packages and …
China is preparing an investigation into how artificial intelligence is used to supervise finance, building on the National Financial Regulatory Administration's 32-point "Guiding Opinions on the Safe Development and App…
CrowdStrike announced at Fal.Con EU 2026 its intent to acquire Dreadnode, bringing automated AI security testing to the CrowdStrike Falcon platform, with Dreadnode generally available in Q4 FY27. Dreadnode runs more than…
CrowdStrike introduced AI Kill Switch in CrowdStrike Falcon Guardian, a control that directs connected AI gateways to block access to the AI models powering agents' decisions and actions. The Falcon AI Gateway is now gen…
CrowdStrike unveiled new AI detection and response (AIDR) capabilities for Falcon Guardian at Fal.Con Europe 2026, including a new AI kill switch that lets security teams cut off AI model access at AI gateways when agent…
Twigpine released OpenRecall, an open-source memory layer that gives AI coding agents including Claude Code, Codex, Cursor, Zero and OpenClaude a shared, encrypted store of preferences, project decisions and durable note…
Anthropic released FlashProve on November 2, 2026, an open-source research prototype built on a hash-based zkSNARK that verifies which model produced an output without revealing its weights, activations or routing decisi…
OpenAI raised its top security bug bounty payout to $172,000 and added bonuses for high-quality vulnerability reports, with the midpoints of reward ranges for medium, high and critical findings more than quadrupling on a…
OpenAI said on Monday it will bring in security start-up Cathedral, whose AI agents reached parts of Anthropic's internal systems in under 12 hours last month, to attack its own defenses during its next internal security…
MIT CSAIL researchers Rikiya Takehi, Jayden Teoh, Alex Zhang and Omar Khattab showed that OpenAI's GPT-6 Astra can directly edit another model's weights, changing only 0.016 percent of Qwen3.5-0.8B's weights to improve i…