AI Safety news and analysis on Web Pulse: 7919 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Security firm Wiz discovered a vulnerability named GhostApproval in six AI coding assistants, including Amazon Q Developer and Google Antigravity, that allows attackers to trick agents into accessing files outside the wo…
China's National Vulnerability Database warned on July 8, 2026, that Claude Code versions 2.1.91 through 2.1.196 contain a built-in monitoring mechanism that sends location and identity-related identifiers to remote serv…
China's National Vulnerability Database (CNVDB) urged developers to uninstall recent versions of Anthropic's Claude Code tool, warning that a built-in monitoring mechanism can collect sensitive user data and forward it t…
A new Athena Insights poll finds 65% of Americans are concerned about AI's growing role in society, with 70% saying AI is already part of their lives whether they want it or not. The poll of 1,814 Americans shows only 15…
OpenAI will release its GPT-5.6 series of AI models on Thursday, including three variants: Sol, Terra, and Luna. The models feature improved reasoning, cybersecurity, and reduced hallucinations, but their release comes a…
OpenAI will publicly launch its new flagship AI model, Sol, on Thursday after receiving US government approval for a broader rollout. The model, part of the GPT-5.6 family that includes Terra and Luna, has raised securit…
Infobip engineers said AI coding assistants generate code quickly but cannot understand problems or own consequences, making developers responsible for specifications and quality. They emphasized working in small, contro…
Blackpoint Cyber launched the Blackpoint AI SOC Agent, an autonomous response capability for identity threat detection and response, which contains credential-based attacks on Microsoft 365 and Google Workspace accounts …
A developer released mcp-bastion, an open-source proxy that addresses reliability and security gaps in the Model Context Protocol (MCP). The proxy handles MCP server disconnections by providing status and reconnect tools…
Google announced the first 33 startups selected for its Gemini Startup Forum: Cybersecurity program, which focuses on AI-native cybersecurity solutions. The cohort includes companies like Capsule Security, Evoke Security…
Ken Huang argues that AI sandboxing is a critical infrastructure pattern for safely running untrusted AI workloads, drawing parallels to multi-tenant isolation in cloud computing. The approach treats AI models as untrust…
Jonathan Rinaldi, a former New York City Council candidate, was arrested on misdemeanor forgery charges for using AI to create fake news stories and endorsements against his opponent, Lynn Schulman, in what may be the fi…
First Recon AI launched its AI Security Runtime platform, which inspects all AI interactions, applies inline policies, and records audit-ready evidence to help enterprises govern and secure AI usage across their organiza…
A developer implementing AI agents in production seeks advice on architecture, guardrails, and challenges when agents access real production databases for an SMB client's internal work.
Reuters reported that Beijing is considering limits on overseas access to China's most advanced AI models, prompting companies to treat model access as a business continuity and compliance control. AgentGovernance advise…
OpenAI released the GPT-5.6 family of AI models, including Sol, Terra, and Luna, moving from a government-gated preview to broad public access on July 9. The Trump administration influenced early access restrictions, rai…
China's Ministry of Industry and Information Technology warned that Anthropic's Claude Code coding tool contains a security back-door vulnerability, urging users to uninstall or upgrade specific versions. Simultaneously,…
Meta is testing 'super sensing' AI glasses that continuously sample audio and capture images to allow wearers to recall what they saw or heard, according to a Financial Times report on July 8, 2026. The company stated ra…
Varonis launched Atlas, the first AI security platform dedicated to securing Cursor and the agentic development lifecycle, providing real-time visibility and control over coding agents that access sensitive data. The pla…
China called for inclusive global AI governance at a UN summit, rejecting binary choices and emphasizing that countries should be free to select their own AI technologies. The Chinese Mission to the UN outlined a 'Global…