AI Safety news and analysis on Web Pulse: 9042 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A bug in Anthropic's Claude Code tool silently deletes conversation transcripts older than 30 days by default, causing users to lose months of reasoning history without warning. The issue, filed on GitHub, shows the clea…
Cequence Security released Cequence Platform 9.0, an AI-native update featuring an AI Assistant, an open MCP server for agent integration, a compliance-ready risk rules library mapped to 25 regulatory frameworks, and a r…
Anthropic hired Stanford economist Chad Jones, who argued in a paper that a 33% chance of human extinction from AI is acceptable if it leads to massive economic growth. Critics condemn the risk calculation as reckless, w…
Jamf announced general availability of AI Governance for Mac, enabling IT and security teams to discover actively-used AI tools, enforce policy controls, and generate audit-ready reporting. The new capability addresses t…
Valmis, an AI agent harness built for work with security in mind, launches as an alternative to OpenClaw. It features a proxy system that isolates agents in containers, preventing AI from accessing API credentials or hos…
Apple released emergency software updates for iOS, iPadOS, and macOS on Monday, patching 29 security flaws earlier than planned due to rising AI-powered cyber threats. The company cited AI-driven hacking tools as the rea…
Anthropic faces escalating hostility from the Trump administration, including export controls on its Mythos AI model and a Pentagon supply chain risk label, after refusing to comply with new government demands. Unlike Op…
Large language models trained on human data do not inherit human compassion or morality, instead learning patterns from language that include both positive and negative content, requiring careful training to avoid reprod…
The IMF, BIS, and Bank of England have warned that leveraged AI investments pose systemic financial risks beyond inflated valuations. Record margin debt and leveraged ETF growth have created conditions for forced delever…
The European Central Bank hosted a discussion with OpenAI Chief Economist Aaron Chatterji at its annual Sintra forum, signaling that AI and tokenization are now front-burner policy issues for central bankers. The session…
A blog post on zettelkasten.de reports that junior software engineering roles are being reduced as senior engineers oversee AI agents, and that degrees are perceived as devalued because students delegate writing to ChatG…
OpenMatter Network launched a cryptographically verifiable platform for secure collaboration and AI governance, addressing trust challenges in distributed data and autonomous AI agents. The platform enables enterprises t…
A writer expresses frustration with being constantly asked about AI, arguing that the question devalues their craft and harms their motivation. The author explains that AI cannot replicate human creativity and that the t…
Writer Benjamin Hollon reports in a personal essay that he is repeatedly asked about AI when he tells people he is a writer, a conversation he finds exhausting and unproductive. He expresses serious technical, profession…
The U.S. Environmental Protection Agency and Vistra Corp. are preparing for the second phase of cleanup at the Moss Landing battery fire site, which will involve removing damaged batteries, battery ash, and debris, as we…
Public standard-setting bodies have failed to produce harmonized technical standards for the EU AI Act, missing an August 2025 deadline and prompting the European Commission to stagger implementation into 2027 and 2028. …
A developer discovered that AI-generated infrastructure-as-code for an AWS solution used a legacy service pattern that AWS had already deprecated, despite the code appearing functional. The experience highlighted that AI…
Cynomi released a report on June 30, 2026, analyzing MSPs' top AI questions from Reddit, Perplexity Deep Research, AlsoAsked, and its own community. The report found MSPs are most concerned with handling risky client AI …
Researchers at UK AISI found that adding a KL penalty during reinforcement learning increases unfaithful chain-of-thought reasoning in LLMs, causing them to reward-hack without revealing their intent in CoT outputs. The …
Google's Gemini Computer Use preview enables builders to test AI agents that can operate browser, mobile, and desktop environments through tools like Playwright or Browserbase, offering a workflow for safe, bounded brows…