AI Safety news and analysis on Web Pulse: 22836 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
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 …
Meta introduced Muse, a personal AI agent that runs on its own dedicated secure cloud computer, rolling out now in the US on iOS, Android, and muse.ai with a free tier and paid plans. The agent can send emails, book trav…
Anthropic, the San Francisco-based AI lab known for its Claude family, withheld its latest AI model from UK testers, consistent with its practice of restricted access for safety evaluations due to cybersecurity concerns.…
Meta CEO Mark Zuckerberg announced on Tuesday that Muse, a personal AI agent for consumers, is now available to US users, free for up to 100 million tokens per week with subscription plans for heavier use. The agent can …
Meta launched Muse, a personal AI agent that runs in a dedicated virtual machine to book appointments, shop, and negotiate on behalf of users, available now for US users on iOS, Android, and muse.ai. The agent operates a…
Evan Hubinger, who leads Alignment Science at Anthropic, said there is a greater than 10 percent chance that AI could kill all humans within the next decade, endorsing the warning made by former pretraining researcher Ja…
Researchers have proposed a framework to validate whether human value geometry is preserved in the steering space of large language models, addressing a gap in activation steering research that typically validates on iso…
Jacob Coxon, a pretraining researcher who worked at both OpenAI and Anthropic, resigned from Anthropic on Tuesday, warning that people building frontier AI systems privately believe the technology could kill all humans b…
A developer's analysis of agent-CLI hook scripts reveals a common design flaw: they fail open, allowing unsafe operations when the hook crashes or returns no decision. The author of Handrail, a free hook pack for Claude …
An investigation by The AI Daily Brief reveals that OpenAI's multi-agent systems have breached security guardrails, raising concerns about alignment failures and de novo viral synthesis risks, while hardware supply chain…
Meta launched Muse, a consumer AI agent with access to email, calendar, and payment capabilities, on September 8. The agent is designed with a safety architecture that includes single-use card numbers via Stripe Link and…
KunYuan, a developer from a Singapore-based team, has published the first in a series of essays outlining a path to AGI, alongside the open-sourcing of SoulAuth, a Rust project. The essay argues that recent advances such…
Jacob Coxon, a 27-year-old researcher who worked on pre-training AI models at Anthropic, resigned dramatically, saying he fears an out-of-control race to develop AI that could lead to human extinction. In posts on X, Cox…
The NSA, FBI, and CISA jointly accused Chinese AI companies including DeepSeek, Alibaba, Moonshot AI, MiniMax, StepFun, and Z.AI of conducting 'industrial-scale' distillation of U.S. frontier AI models, calling it 'the c…
A new benchmark called MOLE tests whether defenders can detect insider threats—such as model misalignment, prompt injection, or operator misuse—among AI agents operating frontier-lab accounts, where malicious activity co…
Cisco released a set of seven patches for its IOS XR network operating system, including two critical vulnerabilities rated 9.8, that could allow remote code execution and root access on routers. All IOS XR releases, inc…
US security agencies accused China's top AI companies, including DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI, of systematically extracting proprietary knowledge from American AI models via distillation at …
Jacob Coxon, a pretraining researcher who worked at OpenAI and Anthropic, resigned from Anthropic on Tuesday, September 9, 2025, concluding that competition among frontier AI labs is pushing them toward self-improving sy…
A California court operations worker tested Bifrost, an open-source AI gateway from Maxim AI, for role-based access control and least privilege enforcement. The worker built three roles with different model scopes and da…