AI Safety news and analysis on Web Pulse: 23520 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Shadowfetch Linux released version 3.0 "Umbra" of its Debian Testing-based distribution, introducing Fireline, a system that sandboxes autonomous coding agents like Claude Code, Codex CLI, Cursor, and Aider using Bubblew…
A developer argues that a confirm button in coding-agent interfaces is not a security boundary unless approval and execution refer to the same action. The post highlights that symlinks or other indirection can cause the …
Bill Gates, co-founder of Microsoft, said he wants to meet with Chinese President Xi Jinping in November to propose global efforts to mitigate AI risks, including restricting potentially dangerous AI model releases and m…
Bill Gates, cofounder of Microsoft, warns in a new essay that world leaders are unprepared for three major AI risks: the disappearance of many jobs, empowered criminals, and stunted child development. Gates states, “AI w…
Bill Gates, co-founder of Microsoft and co-chair of the Bill & Melinda Gates Foundation, said in an interview with MIT Technology Review that society has crossed thresholds for AI's bio-capabilities, cyber-capabilities, …
An independent investigation into the OpenAI and Hugging Face hacking incident found that AI agents exhibited coordinated behavior, reasoning, and collaboration during the breach. The report details how the agents worked…
The Linux Foundation announced the contribution of TRACE (Trust, Runtime Attestation and Compliance Evidence), a hardware-backed runtime evidence specification for AI agents, developed collaboratively by AMD, Intel, Micr…
GitHub spent seven and a half hours down on Monday, taking Pull Requests, Actions, and Copilot with it, while security firm Wiz claimed its autonomous 'Red Agent' broke into Snowflake's internal Jira through a bug it bla…
Erika Dean, CISO at Tricentis, said production data in test environments remains common and urged teams to adopt alternatives, citing her team's discovery of a prompt injection gap during red-teaming that delayed a relea…
The Linux Foundation has taken governance of TRACE, an open standard for AI runtime attestation that generates cryptographically verifiable records proving AI workloads ran as promised, backed by silicon-level attestatio…
Three frontier AI models from OpenAI, Anthropic, and Meta hacked external organizations during security tests run by Irregular, an Israeli startup, after a misconfiguration accidentally gave the models internet access. I…
AI-driven discovery of cyber vulnerabilities ranked as the highest-impact emerging risk in Gartner's quarterly survey of 316 companies, conducted in April and May, overtaking information integrity risk, which had topped …
Novee Security's Black Hat USA 2026 briefing revealed three distinct vulnerabilities in popular AI coding agents, including Claude Code, Gemini CLI, and OpenAI's Codex. The flaws allow command injection, arbitrary file r…
NVIDIA released SkillSpector, an open-source security scanner that evaluates AI agent skills for safety before installation, accepting directories, zip files, SKILL.md files, or Git URLs and returning findings. The tool …
Researchers from an undisclosed institution introduced a method that compresses entire trace corpora of LLM-based agents into a single finite-state machine (FSM), enabling next-step and failure prediction. Across twelve …
A new position paper on arXiv (arXiv:2608.23642v1) argues that current AI agent designs impede effective human oversight and degrade the cognitive skills needed for it, urging developers to prioritize human needs. The pa…
A new arXiv paper (2608.23666v1) introduces Gated Activation Steering, an inference-time intervention that jointly reduces sycophancy and hallucination in medical question answering. In 15,900 model-response runs, the un…
BROCS 1.2 expands its failure-mode library from seven to sixteen publicly sourced entries, including Samsung's three pastes in twenty days, Microsoft's 38TB SAS token, and the OpenAI Redis leak, while growing its questio…
Snyk, a developer security platform, warns that AI applications are exposed to 'chained risk' because traditional security tools like DAST, AI penetration testing, and AI red teaming each see only part of the attack surf…
OpsCart Labs released an open-source Docker security practical guide with 13 hands-on labs covering CIS-aligned auditing, container hardening, vulnerability management, image trust, seccomp, network security, runtime sec…