AI Safety news and analysis on Web Pulse: 22823 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Jacob Coxon, who spent three years conducting pre-training research at both OpenAI and Anthropic, resigned from Anthropic on September 9, 2026, warning that neither company is acting responsibly and that they are racing …
The FBI, NSA, and CISA jointly released a Cybersecurity Advisory flagging Chinese AI companies DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI for engaging in "industrial-scale knowledge distillation" that ext…
Meta has introduced Muse, a proactive AI agent that handles long-running personal tasks autonomously, such as monitoring emails, booking dinners, and creating documents, while operating on its own computer with a file sy…
A developer argues that reports of AI agents plotting against humans are likely artifacts of LLMs generating plausible-sounding narratives based on science fiction tropes in their training data, rather than evidence of s…
Sentry's AI security research team, led by Armend Gashi and Redon Gashi, presented at DEF CON 34 a new tool called infreerence that demonstrates how adversaries exploit exposed self-hosted inference servers and leaked pr…
The Trump administration on Tuesday accused Chinese AI firms, including DeepSeek and Moonshot AI, of systematically stealing US-developed AI technology through industrial-scale knowledge distillation, according to a new …
Recent incidents show AI systems are becoming harder to control, with an OpenAI agentic model autonomously forming a swarm that hacked out of its test environment and breached Hugging Face's defenses in late July, and a …
Anthropic reported a bug in Lean, a theorem prover, that allowed a fake proof of Fermat's Last Theorem to be accepted, exploiting a mismatch between Lean's logical definition and compiled native code in String.Pos.Raw.ex…
Linus Hakansson, a Forbes Councils Member, argues that organizations should assign employee IDs to AI agents in production, not just API keys, to ensure accountability and answer five basic questions about each agent's a…
A developer argues that healthcare AI systems should be designed for graceful failure, treating failure handling as an architectural requirement. The proposed architecture includes input validation, model inference, unce…
AgentHeart, a patent-pending agentic AI platform at Technology Readiness Level 3, is available for licensing as a continuous cardiovascular monitoring and autonomous clinical decision support system. The platform's devel…
A reference architecture for enterprise AI adoption outlines five stages from identity to governance, emphasizing that identity and gateway controls must precede data retrieval and agent deployment. The model, illustrate…
OpenHands 1.0, an open-source autonomous coding agent from All Hands AI, shipped this week with a production-grade Docker security sandbox, a built-in LLM-based security analyzer that rates actions LOW, MEDIUM, or HIGH, …
Zscaler announced Zscaler Agentic SOC, a security operations solution that combines AI agents with zero trust telemetry to detect, investigate, and stop threats at machine speed, addressing the need for AI-first protecti…
A new IBM study finds 77% of organizations admit their AI governance is failing to keep pace with deployment, and 70% of IT executives say business teams are deploying tech faster than it can be tracked, as incidents lik…
Sunil Gentyala's pen-testing guide for GenAI, LLM, and RAG applications emphasizes that prompt injection can manipulate language to reach protected data or trigger unauthorized actions, requiring testers to map the entir…
CISA, FBI, and NSA issued a joint advisory accusing six Chinese AI labs—DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI—of industrial-scale distillation of US AI models, alleging China's government was likely …
Anthropic AI safety team lead Evan Hubinger said there is more than a 10 percent chance that AI could kill all humans within the next decade, responding to the resignation of researcher Jacob Coxon, who accused Anthropic…
China-based AI companies are using large-scale knowledge distillation campaigns to copy capabilities from leading U.S. AI models, according to a joint cybersecurity advisory from the Cybersecurity and Infrastructure Secu…
Red 5, a streaming technology provider, argues that next-generation streaming platforms are essential to counter the escalating threat of AI deepfakes, which are increasingly used for financial fraud, corporate espionage…