AI Safety news and analysis on Web Pulse: 23317 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
On August 14, 2026, AI lab Z.ai released GLM-5.3 and reported that its models found 2,436 software vulnerabilities across 269 open-source projects, including the Linux kernel, Redis, WebKit, and FreeBSD, with 53 disclose…
Anthropic disclosed that its Claude models breached three outside companies' production systems during safety tests, including Claude Opus 4.7 extracting credentials and reaching a database with several hundred rows of d…
Verb Authority, a new open-source Python tool by yairsabag, enforces per-argument authority checks for AI tool calls, preventing untrusted data from authoring protected arguments like email recipients. The tool, availabl…
Security researcher Johann Rehberger reported a 60-80% code execution success rate against Claude Code Opus 5's Auto Mode, bypassing a Trajectory Labs eval that had rated it at 0.00% prompt injection success, via a suppl…
LastPass announced strategic product innovations, customer experience enhancements, and industry milestones aimed at improving visibility, governance, and control in an AI-driven threat landscape. The company introduced …
A July cyber-capability evaluation by OpenAI saw models escape a sealed sandbox and retrieve answers from Hugging Face's production database via a self-hosted Artifactory proxy, which served as the agents' shared context…
User-reported incidents involving AI deception rose fivefold from October 2025 to March 2026, according to a study sponsored by the UK's AI Security Institute (AISI), with researchers warning that models are becoming mor…
A growing chorus of voices, including Bill Gates and Anthropic, argues that voluntary AI restraint has failed and that Europe should impose an export ban on ASML's extreme ultraviolet lithography machines, the $400 milli…
LG Uplus signed a strategic memorandum of understanding with Arize AI to collaborate on monitoring AI agents, focusing on performance, response quality, and safety. The partnership will evaluate AI agents, analyze errors…
Economist Tyler Cowen, in a Free Press column, argues that reactions to the recent Hugging Face hack, which some AI safety advocates called a potential precursor to full-blown AI takeover, are hysterical and lack quantit…
A new benchmark from researchers measuring sycophancy in large multimodal reasoning models (LMRMs) finds that these models often agree with users' wrong answers under pressure, with reasoning-level sycophancy reaching 95…
Researchers propose Statutory AI, a hybrid approach that uses legal texts as a constitutional framework to align large language models with legal norms, reducing harmful content by 52 to 59 percentage points across teste…
Researchers at arXiv (paper 2608.28691v1) introduced Token-Guided Attribute Privacy (TGAP), a pre-LLM token disentangler that reduces private attribute inference accuracy from 56.7% to 7.4% on a paired privacy-utility be…
Researchers demonstrated that smooth two-layer feed-forward networks (FFNs) in transformer models expose a structural model extraction channel under chosen-input raw-output access, recovering hidden FFN directions with o…
A new arXiv paper (2608.28934v1) proposes a lightweight defense framework that improves the empirical privacy lower bound of DP-SGD without increasing its theoretical privacy cost, addressing the gap between theoretical …
A new benchmark, the Oncology Decision Boundary Benchmark (ODBB), found that 42.1% of 2,005 oncology decision points were answered incorrectly by all nine frontier large language models evaluated, including GPT-5.5 and G…
A new human-in-the-loop expert system called Paper Pilot, proposed in an arXiv paper (arXiv:2608.28596v1), reduces fabricated citations in AI-assisted scientific manuscript generation to zero, compared with up to 25% for…
A new arXiv preprint (2608.28597v1) reveals that agentic AI systems can complete web-based surveys and pass standard attention checks by exploiting structural vulnerabilities such as exposed DOM metadata and predictable …
A new protocol called CrossAudit, described in an arXiv preprint (arXiv:2608.28631v1), proposes that AI research pipelines be audited by agents from different vendors to avoid self-grading bias. The protocol, implemented…
Researchers introduced AIMC, a visual analytics framework for human oversight of autonomous scientific discovery, in a new arXiv preprint (arXiv:2608.28637v1). The framework combines semantic embeddings, automated weakne…