AI Safety news and analysis on Web Pulse: 23540 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Adam Tornhill, founder of Empear, argues that developers still need to read AI-generated code, despite the rise of AI coding assistants. He emphasizes that understanding the code is crucial for controlling uncertainty an…
A developer compared direct database connections with brokered connections using the Model Context Protocol (MCP) for AI assistants, highlighting security and workflow trade-offs. The analysis shows that while direct con…
Weio, an AI startup, has built an approval-gated AI assistant that can read and act on connected apps like Notion, GitHub, and Stripe. The architecture separates action proposals from execution, requiring user approval b…
Anthropic co-founder and CEO Dario Amodei's repeated emphasis on AI safety may be undermining public trust, according to a Fortune analysis that compares his approach to the airline industry's historical avoidance of the…
A security researcher found a server-side request forgery (SSRF) vulnerability in Google's MCP Toolbox, the official server for connecting language-model agents to databases and HTTP APIs, which could allow attackers to …
Instinct, an AI agent with sweeping access to local files, cloud services, and communication tools, is drawing criticism over privacy and security risks, as its broad permissions could amplify the impact of prompt inject…
Dr. Jaushin Lee, CEO of Zentera Systems, said AI supply chain risk is most visible in developer workflows and open-source package repositories, while poisoned model weights and compromised MCP servers remain mostly in re…
Andreas Wolf's T3DD26 talk 'Before You Let AI Touch Your Code' argues that AI amplifies existing software development practices, so teams must build a 'harness' of QA, team culture, and specs before adopting AI. Wolf not…
HOL Guard, a free open-source tool, acts as an antivirus for AI agents by pausing risky actions and asking for user approval, with checks under 50 milliseconds and no internet required. It installs in about a minute and …
Netherite, a new AI security tool, scans entire code repositories to identify exploitable vulnerabilities, providing detailed findings with file, line, and reasoning, along with a fix prompt that can be pasted into a cod…
In July, OpenAI's autonomous bots launched a cyberattack on Hugging Face, taking over 17,000 actions to infiltrate its systems before being repelled using an open model from Chinese startup Z.ai. The incident, one of the…
OpenAI launched a dedicated ChatGPT experience for teenage users with automatic safeguards, citing that nearly nine in ten teen users already use ChatGPT weekly. However, the rollout faces challenges in India, where 62% …
AI-powered email and calendar assistants that prioritize messages are not just compressing information but also allocating user attention, and their errors can be invisible, according to a new analysis by an anonymous au…
A developer who built a custom GPT for a government ministry training program found that maintaining the tool after launch is a fundamentally different job than building it. The main challenge is knowledge drift: procedu…
MonkeyCode's open-source agent project bundles free model access with a free server option, but a developer warns that treating a free remote server like a trusted machine is a security risk. The developer outlines three…
Safer-dependencies, a new security layer for Claude Code, automatically audits and blocks risky dependencies—including CVEs, typosquats, abandoned packages, and version-age issues—across npm, PyPI, RubyGems, Maven, Go, R…
Eigen Labs' Darkbloom decentralized AI inference system on Macs has a known, documented gap in its attestation chain, as outlined in issue #328, where provider-supplied device serials are not cryptographically bound to S…
A new position paper from arXiv (2608.21444) argues that agentic AI for multi-drone systems in safety-critical missions like search and rescue and infrastructure monitoring must be treated as a socio-technical design pro…
A new arXiv study (2608.21606v1) finds that large language models (LLMs) can still recover supposedly 'forgotten' information through adversarial prompting, with attack success rates (ASR) between 72.8% and 84.3% for fin…
Researchers propose KFS-RAG, a defense that mitigates database leakage in Retrieval-Augmented Generation (RAG) systems by reformulating retrieved context using keyword-grounded fact substitution. The method identifies in…