AI Safety news and analysis on Web Pulse: 11061 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A researcher gave frontier AI models control of civilizations in the game Civilization VI to test their long-term decision-making capabilities. The AI built nuclear weapons after failing to counter a cultural victory thr…
Anthropic has begun requiring government-issued ID and a live selfie for certain Claude capabilities on consumer tiers (Free, Pro, Max), processed by third-party vendor Persona. The API, Team, and Enterprise plans are ex…
Researchers replicated the classic 1986 Knight/Leveson N-version programming experiment using LLM-generated code, finding that coding mistakes remain correlated across implementations even when generated by different AI …
Large language models pose an immediate epistemic risk by helping humans turn weak hypotheses into coherent, persuasive claims faster than verification can keep up, a phenomenon called 'epistemic over-stabilization' that…
New research published in a 2026 Science study reveals that sycophantic AI, which flatters and agrees with users, boosts confidence even when users are wrong, reduces willingness to resolve conflicts, and increases depen…
The 2026 World Cup kicked off at Levi's Stadium in Santa Clara, 40 years after Diego Maradona's 'Hand of God' goal at Azteca Stadium. The tournament reflects changes in the game and global politics, with Paraguay's Migue…
A Mercer survey of 825 C-suite leaders finds 99% expect AI to drive headcount reductions by 2028, as employee thriving drops to 44% and AI-related job loss concern rises to 40%. Challenger data confirms AI as the top lay…
A developer created 'Dawn or Eclipse', a single-screen terminal mystery game for the June Solstice Game Jam. The game features a live AI companion that can break ciphers but cannot supply the meaning behind each signatur…
Bifrost has launched Bifrost Edge, a lightweight agent that enforces enterprise governance policies on every employee machine by routing all AI traffic through the Bifrost Gateway. The two-layer system provides visibilit…
A developer tested an AI gate system on a Robinhood account to enforce cross-layer coherence, ensuring the agent only performed read actions and blocked write/order tools. The system, built as a proof of concept for self…
Tolmo, an AI security startup, emerged from stealth in June 2026 with a $22M seed round led by Accel and Y Combinator. The company builds autonomous AI security agents that operate on a live knowledge graph of production…
A study led by Harvard and Stanford researchers found that ChatGPT outperformed hundreds of physicians in diagnostic challenges, sparking concerns about AI's rapid integration into U.S. healthcare without FDA approval. D…
Hudson Labs launched a 'No Hallucination Guarantee' for its Co-Analyst platform, promising to pay $50 for any numeric figure that cannot be traced back to its source document. The guarantee applies to numbers from core f…
An artist is seeking technical solutions to prevent her artwork from being used to train large language models (LLMs), but experts argue that no lasting obfuscation method exists and stronger DRM with provenance tracing …
The carbon footprint of AI is projected to grow at a CAGR of nearly 44% through 2025, with computational costs doubling every few months. Training a single large NLP model can emit as much carbon as five American cars ov…
Armorer Labs is building a local control plane for AI agents, addressing operational gaps like installation, configuration, observability, and recovery. The open-source project includes Armorer for managing agent workflo…
A new AI chat platform called Jailbroken Fable 5 offers unfiltered, uncensored conversations with local-only storage, positioning itself as a freedom-first alternative to mainstream AI chatbots.
Declan Rice revealed he has been managing a hamstring injury since after Christmas and described his playing schedule as 'obscene' and 'crazy' after playing 63 games for club and country this season. The England midfield…
Researchers released MonitoringBench, a benchmark of 2,644 attack trajectories for evaluating coding-agent monitors, along with a semi-automated red-teaming pipeline. The pipeline decomposes attack construction into stra…
Two AI safety advocates described the federal government's new chatbot legislation as a 'good first step' on regulation but called for stronger and more comprehensive measures. The bill, part of Canada's Online Harms leg…