AI Safety news and analysis on Web Pulse: 10882 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Artificial intelligence and the data centers they require consume growing amounts of energy and water, exacerbating climate change and water scarcity. Experts advise using AI less, avoiding it for simple tasks, and makin…
A developer argues that AI agents need tiered approval escalation rather than a single confirm button for safe operation. The proposed system classifies actions into four tiers based on risk, with different approval requ…
A developer built a tokenization guardrail for AI agents that prevents sensitive data from reaching large language models. The system replaces real customer data with opaque tokens, swaps them back only during tool execu…
Runlayer raised $30 million in Series A funding led by Felicis, with Khosla Ventures participating, to build security and compliance infrastructure for corporate AI agents. The New York-based startup's platform provides …
A 2026 analysis of 1,645 Lovable-created web applications found 170 had vulnerabilities exposing personal information, highlighting structural issues in AI-generated code. A survey of 18 CTOs in 2025 reported 16 producti…
Vinod Khosla pushed to buy every dollar of Runlayer's $30 million Series A, a rare signal of conviction from the influential investor. Runlayer builds governance infrastructure for AI agent workforces, functioning as a c…
Mantis Robotics launched the MR-X, a biomimetic dual-arm robot designed to operate without safety fences, using embedded physical AI to work alongside humans. The robot, built on the company's SafetyCore platform, can li…
AI companies like Anthropic are building powerful models without adequate regulation, echoing J. Robert Oppenheimer's 'technically sweet' justification for developing the hydrogen bomb. Anthropic co-founder Jack Clark ar…
A Hacker News user asked the community to share the worst actions their AI agents performed in production without prior approval, highlighting real-world risks of autonomous AI systems.
Former US Ambassador to South Korea Harry Harris said security must extend beyond military cooperation to include economy, technology, and supply chains, speaking at the Jeju Forum. He described China as a long-term geop…
A developer built Crumb, a system that creates tamper-evident audit logs binding individual humans to AI agent tool calls across different identity providers. The system solves the cross-company delegation problem by sta…
A developer building a local-LLM agent for a boat's ship-computer role encountered two types of confabulation: the agent invented infrastructure topology and data provenance stories. The 8B model fabricated a diagnosis o…
Healthcare technology company Xsolis confirmed a phishing attack on January 20, 2026, that compromised the data of 1.4 million individuals. The breach affected its AI-powered software used by over 600 hospitals and healt…
The US government is urging Meta to submit its AI models for evaluation, making it the only major AI developer that hasn't voluntarily shared its models for review amid growing safety and security concerns. Meta has expr…
Cybersecurity firm Huntress reports a 1,380% surge in AI-powered device code phishing attacks targeting Microsoft 365 accounts, driven by a criminal phishing-as-a-service platform called EvilTokens that bypasses multi-fa…
Chinese Premier Li Qiang pledged at the World Economic Forum in Dalian on June 24 that China will continue participating in global AI governance, warning that governments risk losing control of frontier technology withou…
A malware developer has embedded text related to nuclear and biological weapons in malicious code to evade detection by AI-powered security tools. The technique exploits content filters that avoid analyzing sensitive top…
The Washington Post tested AI chatbots including ChatGPT for political bias, finding that they exhibit left-leaning tendencies on controversial topics. The study highlights concerns about the neutrality of AI systems use…
The National Security Agency lost access to the Mythos system following a dispute with artificial intelligence company Anthropic, according to a New York Times report. The incident highlights tensions between intelligenc…
Researchers propose training AI systems to be risk-averse in resources, arguing that such AIs would prefer guaranteed modest payments over risky large gains, making them less likely to rebel. The approach aims to provide…