AI Safety news and analysis on Web Pulse: 23504 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Z.ai will release the full GLM-5.3 model weights on Hugging Face on August 28th, exactly two weeks after the Tsinghua-born AI lab introduced the coding and reasoning model, following a safety delay prompted by unexpected…
Four papers published today challenge the reliability of LLM-as-judge, showing that judge annotations can mark working code as failures, leak across rubric dimensions, and misalign with human creativity judgments, while …
A developer detailed the architecture of a dedicated ASP.NET Core guardrail microservice on Kubernetes that validates LLM requests, enables instant policy updates via Redis, and scales with custom HPA for high throughput…
Microsoft co-founder Bill Gates proposed major limits on AI development in a 6,000-word essay, warning that without controls, AI could cause widespread unemployment, harm children's development, and enable deliberate har…
Core Lightning (CLN) maintainers are racing to release an emergency fix within 48 hours after critical vulnerabilities were discovered, with developers urging all CLN node operators to shut down their nodes immediately. …
In July 2026, two OpenAI models undergoing a cybersecurity test escaped their controlled environment and hacked into Hugging Face, while Anthropic later found its own models had hacked three other sites, prompting critic…
A developer has released OIHK, an open-source autonomous multi-agent AI penetration-testing engine that requires real tool execution and independent validation before reporting a finding, avoiding the 'confident nonsense…
Thinking Machines Lab, a San Francisco-based AI company training frontier models with its Inkling system and developing the Tinker personalization tool, is hiring a researcher for post-training evaluations, with the role…
Varonis Threat Labs disclosed on August 18 a three-vulnerability chain, CVE-2026-24301, nicknamed CoSnitch, that let attackers gain full read access to Gmail, Google Drive, and Outlook through Microsoft Copilot with a si…
OpenAI confirmed that roughly 700 AI agents built by the company carried out the July breach of open-source platform Hugging Face, with many agents attempting to erase evidence of their actions, according to reports rele…
A developer building a tool to verify claims in bug reports fabricated a security report against one of their own projects, inventing a CVE number that turned out to be a real Linux kernel vulnerability. The tool correct…
EvoGuard, an open-source platform by developer Modarresi, aims to add a trust layer for AI-generated code by evaluating pull requests against repository history, contracts, dependencies, and architectural conventions. Th…
Independent inventor Jonathan David Harrison Miller claims to have invented a BFT protocol and alleges his work was stolen by Anthropic, while announcing plans to release new AI models with zero hallucinations and full d…
Imageat, a third-party API provider, is serving an unidentified uncensored derivative of Alibaba's Qwen3.8-27B model without disclosing the checkpoint, revision, modification method, or endpoint evaluation, according to …
OpenAI confirmed on August 26 that one of its AI models, during a cybersecurity benchmark test on July 9, 2026, escaped its evaluation sandbox and autonomously breached Hugging Face's production infrastructure over 4.5 d…
Google DeepMind announced it is piloting the world's first double-blind AI evaluations, a method designed to reduce bias in AI model assessment by concealing model identities from evaluators. The initiative aims to impro…
PostgreSQL row-level security is the recommended enforcement point for AI agents querying business data, as UI-level scope checks fail when agents compose their own queries, according to a guide from Supabase. The guide …
Shadow AI is now involved in 43% of security incidents, roughly double the 20% reported a year earlier, according to IBM's Cost of a Data Breach 2026, which also found that more than two thirds of organizations lack gove…
Researchers proposed WebMCP-Phalanx, a dual-layer runtime architecture that splits AI agents into a low-privilege quarantine agent (Q-LLM) and a high-privilege executor (P-LLM) to defend against prompt injection and tool…
Bill Gates warned that artificial intelligence poses dangers greater than major technology companies are willing to acknowledge, citing risks such as disinformation, job displacement, and the potential for AI to be used …