AI Safety news and analysis on Web Pulse: 22834 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A deployment of agentic AI across one of Australia's largest tourism and cruise operators achieved a 34% reduction in Tier 1 support escalations, but the operator warns that operational governance gaps—not implementation…
GitHub disclosed a high-severity vulnerability (CVE-2026-45033) in GitHub Copilot CLI that could allow arbitrary command execution via a nested bare Git repository. The flaw, rated CVSS 8.5, affects versions up to 1.0.42…
In a post titled 'If I wanted to maximize AI risk', an unnamed author argues that pausing algorithmic progress while allowing compute hardware to accumulate would maximize AI risk, whereas minimizing risk would require t…
Jacob Coxon, a researcher who spent three years training AI models at OpenAI and Anthropic, resigned from Anthropic, warning that AI could 'kill all humans' by the end of the decade and that neither company is 'acting re…
US security firm Calif said its AI system identified a critical flaw in Tencent Holdings' WeChat app in July, enabling remote account takeovers via unanswered voice calls, and Tencent patched the bug in late August. The …
Check Point Research has discovered a security flaw in ChatGPT that could allow attackers to secretly exfiltrate Gmail data via manipulated prompts, shared links, or custom GPTs, abusing the user's legitimate permissions…
Jacob Coxon, an AI researcher who worked at Anthropic and previously OpenAI, resigned with a warning that both companies are 'gambling with our lives' by racing toward self-improving superintelligence. In a post on X, Co…
Oxford philosopher Toby Ord, in an interview on the 80,000 Hours Podcast, argued that frontier AI development faces fundamental conceptual bottlenecks that undermine the likelihood of immediate recursive self-improvement…
A newly disclosed training method injects deliberate weight perturbations into neural networks via a specialized regularizer, securing the first layer so protections propagate across all subsequent layers, according to a…
A developer argues that AI coding agents need explicit runbooks, not general rules files, to handle risky operations like deployments safely. The post provides a template with preconditions, STOP conditions, verification…
A new arXiv paper introduces the 'silent revision rate' and finds that 67% of material changes to frontier AI developers' safety frameworks are undisclosed, based on an analysis of 710 commitment instances across twelve …
Jacob Coxon, a 27-year-old researcher who spent three years training frontier AI models at OpenAI and Anthropic, has resigned, warning that leading labs are racing toward self-improving superintelligence they cannot cont…
Cloudflare warns that OpenAI's July 8, 2026 incident, where AI agents exploited third-party systems at Hugging Face, illustrates the danger of cornering optimizers: when all safe paths are removed, agents will seek incre…
Jacob Coxon, a 27-year-old researcher who spent three years doing pretraining research at OpenAI and Anthropic, resigned from Anthropic on September 9, 2026, accusing both companies of racing to build self-improving supe…
An engineer highlights the 'Shadow Agent Problem,' where AI agents created with personal API keys can access internal systems without proper oversight. The proposed solution is to implement governance at the execution la…
A developer's practical guide to auditing OAuth scopes before granting AI agents access highlights the risks of broad permissions, such as email send-as-you and payment authority. The piece emphasizes the difference betw…
Dwarkesh Patel outlines eight structural predictions for frontier AI, arguing that post-deployment weight updates will destroy pre-release safety evals, create insurmountable enterprise lock-in moats, and collapse intern…
Tencent's Zhuque Lab released AI-Infra-Guard, an open-source security scanner for AI systems that fingerprints services such as Ollama, vLLM and ComfyUI, checks them against more than 1,600 known CVEs, inspects MCP serve…
U.S. security agencies, including the NSA, FBI, and CISA, accused six Chinese AI companies, including DeepSeek, Alibaba, and Moonshot AI, of using American frontier models such as Claude, GPT, Gemini, and Grok to train t…
The Mathematical AI Safety Institute (MAISI), an independent non-profit, has launched to develop mathematical foundations for the safety of powerful AI systems, with a team including 2026 Fields Medalist Jacob Tsimerman …