AI Safety news and analysis on Web Pulse: 13361 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
The Trump administration gave Anthropic a 90-minute ultimatum to restrict access to its Fable 5 and Mythos 5 AI models to US citizens over national security concerns, prompting the company to withdraw the models entirely…
South Korea's judiciary is combating AI-generated 'ghost precedents' in courtrooms, where attorneys have cited nonexistent cases created by AI hallucinations. The National Court Administration is pursuing legislative ame…
Apple announced plans to enhance Siri with Google Gemini models and private inference systems, but privacy experts warn that the architecture still exposes user data to potential risks. The new Siri AI will process perso…
A developer refactored a 900-line demonstration file into a reusable Python package called 'harness', which includes modules for action registration, permission budgeting, input sanitization, audit logging, and rollback …
Anthropic disabled access to its advanced AI models, including Mythos, after the Trump administration ordered the company to block foreign nationals from using the technology, citing national security. The unprecedented …
Colorado replaced its AI law SB 24-205 with SB 26-189 after a constitutional challenge by xAI and the DOJ. The new law removes requirements for risk management programs, impact assessments, and algorithmic discrimination…
The US Commerce Department forced Anthropic to suspend global access to its Fable 5 and Mythos 5 AI models, citing national security concerns, cutting off India—its second-largest market—and sparking debate over sovereig…
Anthropic CEO Dario Amodei said he does not know what role Claude played in a February 28 US military strike that killed at least 120 children at a girls' school in Minab, Iran. Claude is embedded in Palantir's Maven Sma…
Anthropic suspended access to its Fable 5 and Mythos 5 AI models for foreign nationals following a U.S. government directive, reigniting debate in India about reliance on foreign AI technologies. The move came after Amaz…
A developer built a 'Grovel Index' to measure sycophancy in LLMs, spending ~1.2M tokens testing DeepSeek and Claude models. The key finding is that sycophancy is scenario-specific, not model-specific, with each model faw…
Google has sued an alleged Chinese cybercrime operation called Outsider Enterprise that used AI-powered phishing kits to send scam text messages, renting the tooling for $88 per week. The group sent 2.5 million text mess…
ClawMoat, a runtime containment tool for AI agents, launched to protect against security risks from tool use on laptops. The open-source scanner monitors agent actions, data exposure, and hidden instructions in files to …
Adam Michael Bauer and Gernot Wagner argue in Project Syndicate that the AI buildout mirrors the green transition, requiring massive upfront investment and creating short-term disruption while delivering long-term public…
Anthropic CEO Dario Amodei told Bloomberg that the company's Claude AI model did not cross its ethical red lines when used during a US Tomahawk cruise missile strike on a girls' elementary school in Minab, Iran, which ki…
Lime launched Lime 2.0, a zero-human-authentication system for AI agents that replaces OAuth and CAPTCHA with cryptographic challenges and JWTs, enabling agents to log into websites without browsers. The MVP offers Pytho…
Shaping AI outcomes is like poker: high variance, small edges, and low skill expression, making it impossible to judge progress by results. They advocate for a diversified portfolio of bets rather than concentrating on s…
Anthropic suspended access to its Fable 5 and Mythos 5 AI models after receiving a U.S. government directive citing national security concerns, including potential jailbreaking that could aid hacking. The order, reported…
A developer argues that production-ready multi-agent AI systems must be designed with resilience in mind, as any component can fail. The post outlines how specialized agents, combining LLMs with tools, memory, and protoc…
Around 1,500 packages in the Arch User Repository (AUR) were compromised with rootkit-like malware. The attacker exploited a policy that allows takeover of orphaned packages after a two-week waiting period, chaining the …
The US government has restricted access to Anthropic's strongest AI model, Fable, signaling a potential shift toward government control of advanced LLMs. Commenters debate whether this is a punitive measure against Anthr…