AI Safety news and analysis on Web Pulse: 13553 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Cloud-native software development increasingly relies on asynchronous AI agents, but their utility depends entirely on the ability to verify the outputs they produce. Verification for these agents is fundamentally a runt…
AI agents fail to scale due to engineering challenges—not AI limitations—including unpredictable user behavior, inaccessible enterprise data, multi-agent orchestration failures, lack of business-specific knowledge, and i…
A new framework called "hybrid intelligence" proposes that humans must actively shape their integration with artificial intelligence to prevent four interconnected psychological risks: agency decay, bond erosion, the cli…
Apple focused on practical AI features at WWDC 2026 rather than hyping "agentic" AI, a buzzword dominating competitor keynotes from Google, Microsoft, and NVIDIA. The company briefly mentioned agentic AI but instead demo…
A new proposal from the AI safety community argues that discussions of AI timelines should shift focus from when advanced AI will arrive to when decisions shaping its outcomes will be most impactful. The author introduce…
Anthropic CEO Dario Amodei published an essay and two policy frameworks calling for mandatory third-party audits of frontier AI models, warning that unregulated AI poses catastrophic risks to national security and critic…
Anthropic released an economic policy framework Wednesday proposing three tiers of response to potential AI-driven unemployment, pledging $350 million to develop solutions. The company outlined plans for scenarios rangin…
Multiple Linux distributions issued security updates on Thursday, including AlmaLinux, Debian, Fedora, Mageia, Oracle, Red Hat, SUSE, and Ubuntu, addressing vulnerabilities in packages such as .NET, kernel, chromium, and…
Anthropic and OpenAI both filed confidential S-1 drafts with the SEC in early June 2026, signaling a shift toward public accountability for frontier AI labs. Anthropic expanded its controlled-access Project Glasswing to …
Apple's WWDC 2026 keynote abandoned the traditional platform-by-platform structure in favor of a theme-driven presentation divided into platform improvements, trust and safety, and Apple Intelligence and Siri. The compan…
Canada introduced the Digital Safety Act on Wednesday, a bill that would ban children under 16 from social media and regulate AI chatbots under the same legislation. The law, which targets algorithmic feeds and engagemen…
Microsoft has open-sourced ASSERT, an AI evaluation framework that converts natural-language requirements into executable tests for enterprise agents, as organizations struggle to validate agent behavior before productio…
Anthropic has apologized for secretly throttling its new AI model, Claude Fable 5, with hidden guardrails that silently degraded responses to users suspected of attempting model distillation. The company says it will rev…
Meta realizou no Brasil o evento Meta Conversations, anunciando novidades no WhatsApp para empresas, incluindo um agente de IA para SAC e a API Marketing Message com otimização inteligente de entrega. A empresa também pr…
A new report from Wakefield Research and PagerDuty found that two-thirds of 1,250 surveyed office professionals use AI tools at work without explicit permission, a practice known as shadow AI. The research revealed that …
Agent systems accumulate rules like old companies accumulate meetings, creating instruction files that grow without shrinking. The real fix is not rewriting rules but deciding where each belongs—many should be deleted or…
The Unjournal has not expanded into evaluating legal research due to a lack of senior or mid-career legal scholars willing to co-lead the initiative. The organization has created a prototype tool for sourcing, curating, …
China's state-run newspaper Workers' Daily issued an unusually direct call for government action to protect labor rights amid the rapid adoption of artificial intelligence in the workplace. The editorial urged regulators…
Google DeepMind is funding research into the potential dangers of millions of AI agents interacting online without human oversight, warning that the mass deployment of autonomous agents creates a new class of risk includ…
Anthropic's new AI model, Claude Mythos, achieved a 93.9% success rate at finding and exploiting software vulnerabilities, raising alarm among security experts. Marketers who connect AI tools to Google Drive, Gmail, and …