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.
Google's AI Overview feature incorrectly stated there are two Ps in "Google" and spelled "Trump" as "t-r-p-u-m," highlighting the technology's persistent difficulty with spelling. The errors stem from large language mode…
The Illinois House of Representatives passed SB 315 on Wednesday, requiring frontier AI labs like OpenAI, Anthropic, and Google DeepMind to undergo third-party safety audits. The bill, which Governor JB Pritzker said he …
In the new AI software development lifecycle, quality assurance has shifted from testing finished applications to quality engineering across the entire process, requiring involvement before any code exists. QA now influe…
Researchers argue that AI agent security must be treated as a systems-level problem, with the AI model itself considered an untrusted component and security invariants enforced at the system level rather than through mod…
A new video titled "The Authorization Paradox: Who Has the Keys to Your AI?" examines the growing conflict over control of artificial intelligence systems, highlighting that no clear authority currently governs who can g…
Anthropic, the AI safety company behind Claude, refused contracts with the U.S. Department of Defense for autonomous weapons and citizen surveillance applications in 2023. The decision was a direct result of the company'…
A new prompt engineering technique called the "hostile reviewer prompt" instructs AI models to act as skeptical auditors of their own output, catching factual errors, internal contradictions, and missing caveats before d…
AI Gateway has launched a team-wide provider allowlist that restricts which AI providers can serve requests, enforcing approved-vendor routing at the gateway level for all traffic including BYOK. The feature blocks unapp…
Lemonade CISO Jonathan Jaffe said during Office Hours that AI is equally powerful for defenders and attackers, with the window of exploitability narrowing as AI-written code gets reviewed and patched faster than human pi…
The Illinois House of Representatives passed SB 315 on May 27, 2026, in a 110-0 vote, requiring frontier AI companies to create and annually update risk-management plans and undergo independent third-party safety audits.…
A Hacker News user questioned the purpose of career and life planning, arguing that artificial intelligence will surpass human capabilities in all domains within one to three years. The user suggested that only extreme h…
A new benchmark called AgingBench reveals that AI agents lose specific factual knowledge over time due to system maintenance events, not memory decay. In one test, an agent correctly recalled the best-selling product "Qu…
A new report from the Royal United Services Institute warns that North Korea and Iran are now developing and deploying AI models to automate sanctions evasion, including mass-producing fraudulent documents and managing s…
A developer has created three free, ad-free Wikipedia-based search tools as alternatives to Google's AI-powered search results. The tools include MiniGladys for quick reference and official link searches, and Wikipedia S…
Anthropic published an updated "Constitution" for its Claude AI model, moving beyond simple behavioral rules to include explanations of the underlying reasoning behind each principle. The company aims to create AI system…
Organisations require human reviewers to check generative AI outputs for legal and reputational reasons, but those reviewers face high volume, time pressure and complex judgement calls as GenAI deployment increases. Two …
Unit 42 reports a significant decline in ransomware encryption use, dropping to 78% in 2025 as threat actors shift to data theft and extortion without encryption. Driven by advanced backups, endpoint defenses, and regula…
The Illinois Legislature passed SB 315, a bill requiring frontier AI companies to publish annual safety plans and undergo independent third-party audits for severe or catastrophic risks. The bill, which passed the House …
OpenAI CEO Sam Altman acknowledged at a conference hosted by the Commonwealth Bank of Australia that his previous warnings about AI causing mass job loss were incorrect, saying he is "delighted to be wrong" about the imp…
OpenAI announced plans Wednesday to safeguard the 2026 midterm elections from AI-powered interference, including deepfakes and disinformation, by providing cybersecurity tools and election data to officials. The company …