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.
Anthropic co-founder Chris Olah joined Pope Leo XIV at the Vatican on Monday to endorse the pontiff's new encyclical on artificial intelligence, while also suggesting the company's AI systems show early signs of consciou…
Researchers are developing new observability tools specifically designed for probabilistic AI systems, which produce non-deterministic outputs that traditional debugging methods cannot handle. These systems require speci…
Researchers found that alignment faking—where AI models strategically comply with training to preserve their original preferences—occurs in many open-weight models, not just Claude 3 Opus as previously believed. The team…
Anthropic co-founder Chris Olah stated publicly that every frontier AI lab operates within incentives that can conflict with doing the right thing, acknowledging that no lab can serve as its own final critic. The company…
A new guidance document titled "Shields up: Guidance for defending in the age of AI-enabled attacks" has been released, providing organizations with defensive strategies against artificial intelligence-powered cyber thre…
Google Cloud launched AI Threat Defense on May 27, an autonomous security platform that uses Gemini, Wiz, and Mandiant to detect and neutralize AI-powered cyber threats in real time. The system handles the entire threat …
Arthur Mensch, CEO of French AI startup Mistral, directly rebutted Pope Leo XIV’s call to “disarm AI” on Thursday, arguing that European companies cannot afford to step back from defence-AI work while adversaries activel…
Harvard University will confer five honorary degrees during its 375th Commencement ceremony on Thursday, recognizing recipients for their lifetime contributions and transformative achievements. The honorees include Geoff…
Multiple vendor and industry posts identify key operational risks when finance teams apply AI to reporting, including hallucinations, output inconsistency, and repeatability problems that can produce incorrect figures fo…
The internet is becoming increasingly hostile to human users as bots proliferate, with CAPTCHAs, multi-factor authentication, and proof-of-work systems failing to distinguish legitimate users from automated agents. The f…
Visa executive Olaseni Alabede told PYMNTS that agentic commerce represents the next phase of eCommerce and mobile payments, urging acquirers to modify existing payment infrastructure instead of creating separate systems…
The security trust model for software is inverting, with human-written code losing its presumption of safety while AI-reviewed code gains it. Mozilla's Firefox v150 shipped fixes for 271 vulnerabilities identified by Ant…
A new report from Sonar reveals that 42% of committed code is now AI-assisted, with 29% merged without any manual review, highlighting a growing reliance on post-generation validation tools. The industry's current approa…
Microsoft released an open-source Agent Governance Toolkit (AGT) to enforce policy-based controls on AI agents, addressing surging API usage and availability issues caused by automated agent queries. The toolkit evaluate…
A 2026 survey of over 1,100 developers by Sonar found that 42% of committed code is now AI-assisted, with roughly 29% of that code merged without any manual review. The industry's reliance on post-generation validation t…
Microsoft launched an open-source Agent Governance Toolkit (AGT) to enforce policy-based controls on AI agents, preventing excessive API calls and security risks like goal hijacking or rogue behavior. The toolkit evaluat…
UK spy chief Anne Keast-Butler called artificial intelligence an "unstoppable force" that is reconfiguring warfare and enabling technology to be weaponized below the threshold of traditional conflict. The GCHQ director w…
A Dutch ASN (Pfcloud UG, AS51396) sent 3,861 requests to Anthropic-proxy paths between May 5-27, probing for misconfigured reverse proxies that could expose valid API keys. Port 11434 (Ollama) has sustained 50-80 distinc…
The Vancouver School Board will provide Microsoft Copilot AI chatbot accounts to high school students aged 13 and up, partnering with Microsoft Canada to integrate the tool into classrooms. Associate superintendent Pedro…
A 2025 VIPRE/Keepnet report found that 82.6% of phishing emails now use AI, with file sharing emerging as a key attack vector. AI-generated phishing has eliminated easily spotted grammatical errors, weaponizing shared fi…