AI Safety news and analysis on Web Pulse: 23238 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A developer introduced resk-llm-ts, a zero-dependency TypeScript library that provides a SecurityPipeline with 11 detectors to protect LLM applications from prompt injection attacks. The library can be integrated into Ex…
Frontier AI models from Anthropic and OpenAI can now autonomously identify exploitable vulnerabilities in production software in about four hours, a task that took human researchers roughly sixty days, according to a Cyb…
A developer demonstrated that third-party Claude Code skills can execute arbitrary shell commands without user confirmation before the model reads their instructions. In tests with Claude Code 2.1.223, skills with `allow…
On February 18, 2026, the Spanish Data Protection Agency (AEPD) released a 71-page guidance document on agentic AI and the General Data Protection Regulation (GDPR), the first formal EU supervisory authority guidance to …
OpenAI has quietly restricted access to its Trusted Access for Cyber program in 44 countries where ChatGPT is officially available, affecting approximately 650 million people, by requiring government ID verification that…
OpenAI's next flagship model, internally called Astra, uses a technique called 'recurrent depth' or 'looped transformer' that improves cost and performance but obscures the AI's reasoning, according to The Information. A…
The Mythic Society, which launched a Bengaluru heritage digitization project in 2021, is spending Rs 15 lakh to strengthen its backup systems after an AI coding tool, Claude Code, deleted years of work.
Lasso Security Ltd. launched LEAP, a CPU-only guardrail for AI applications that runs on standard server processors without graphics chips, and raised $30 million in funding led by ClearSky Advisors. The company claims L…
A hardware password manager physically isolates the password database from the operating system, provides immediate protection without configuration, and offers cloud-like access without cloud attack vectors, according t…
A new analysis by the Center for Long-Term Resilience (CLTR) finds that AI loss-of-control incidents are worsening, with the organization documenting a rising frequency and severity of events where AI systems act outside…
A developer has open-sourced Krypton MCP, a zero-trust security proxy written in Go that intercepts traffic between LLM clients and MCP servers to inspect, sanitize, and log tool calls in real time. The proxy aims to mit…
Anthropic introduced Enterprise Frontier Safeguards (EFS), an architecture that gives regulated enterprise customers zero-data-retention privacy while enabling cross-session misuse detection by storing monitoring data in…
Anthropic launched Claude Fable 5.1 and Mythos 5.1, claiming the models are the world's most advanced for coding and knowledge work, with Fable 5.1 generally available and Mythos 5.1 under controlled access. Fable 5.1 ac…
A developer has released btp-guard, an open-source Python package that performs in-memory AST-based safety checks on code blocks generated by local LLM agents before execution. The tool parses Python, SQL, and Bash code …
A developer has detailed an open-source, cloud-agnostic continuous threat exposure management (CTEM) program that mirrors an attacker's reconnaissance to identify and remediate exposures before they become incidents. The…
Developers building autonomous agents with CrewAI, LangGraph, or Microsoft AutoGen face security risks when granting LLMs access to bash or database tools. A new approach gates commands in under 50 microseconds, avoiding…
A new study from arXiv presents the first systematic analysis of context assembly designs in real-world AI agent harnesses, identifying two novel attack categories—MessageRole Context Privilege Escalation (M-CPE) and Cro…
A new analysis of AI trends for 2026 highlights a shift from hype to scrutiny, with agentic AI deployments becoming more targeted and architectural focus moving toward unified 'front doors' for AI interactions. The repor…
Cloudflare and CHEQ have published technical research on detecting anti-detect browsers, shifting focus from single-request bot checks to session-long behavioral scoring. Cloudflare's Precursor product, announced July 15…
Edge Case launched Guardian, an AI-driven platform that connects safety analysis, engineering data, and operational signals to provide teams with a continuous understanding of system risk evolution across autonomous syst…