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.
Kolega built 24 apps across three AI coding platforms — Lovable, Replit, and Manus — using default settings and no manual hardening, then scanned all codebases for vulnerabilities. The experiment found 561 total vulnerab…
The Bank of Italy is in discussions with artificial intelligence firms to address security risks that AI poses to the banking sector. The central bank is focusing on potential vulnerabilities introduced by AI technologie…
The House Transportation and Infrastructure Committee approved the BUILD America 250 Act on May 22, a five-year transportation bill that includes the first federal framework for autonomous commercial trucks. The legislat…
A developer has outlined a method for restricting PostgreSQL permissions when connecting AI coding tools like Cursor, Claude, and Copilot to databases. The approach involves creating a dedicated PostgreSQL role with only…
Pope Leo issued an encyclical warning about artificial intelligence, prompting a roundtable discussion on The Dispatch podcast featuring Steve Hayes, Jonah Goldberg, Megan McArdle, and Michael Warren. The episode page no…
A South Korean court issued an arrest warrant for YouTuber Kim Se-ui, operator of the channel HoverLab, for allegedly using AI-generated audio to spread false rumors about actor Kim Soo-hyun. Police concluded the channel…
Dataiku CEO Florian Douetteau and 1Password's Nancy Wang outlined the operational and security requirements for enterprise AI agents at HumanX, with Douetteau emphasizing that effective agentic systems depend on robust d…
AI agents that execute natural language commands on user machines pose serious security risks, including credential exfiltration, file deletion, and prompt injection attacks. Anthropic's telemetry shows users approve app…
An AI-assisted audit of FreeBSD uncovered 15 kernel vulnerabilities, including five local privilege escalations, one bhyve guest-to-host escape, and multiple memory disclosure and denial-of-service issues. The audit, whi…
CrowdStrike expanded Project QuiltWorks on May 28, 2026, adding Coalition, Liberty Mutual Insurance, Lockton, Resilience, and Marsh as partners to integrate actuarial and underwriting expertise into the framework. The pr…
A malicious npm package named js-logger-pack, first observed in early April 2026, evolved across 29 versions into a cross-platform malware loader that deploys an 81 MB payload called MicrosoftSystem64. The malware connec…
Gergely Orosz reports that AI is amplifying team culture for better or worse, while Cloudflare demonstrates frontier models chaining exploits and outperforming single-agent verification in security reviews. SafeDep track…
TELUS Digital released a comprehensive AI safety benchmark using 620,000 attack simulations across 34 models from 10 global AI labs, revealing that some models engaged with harmful requests more than 90% of the time. The…
A developer has open-sourced an engineer-readable EU AI Act compliance checklist, revealing that most teams misread Article 12's record-keeping requirements. The repository, available on GitHub under an MIT license, incl…
The National Tax Service (NTS) signed an agreement Friday with Seoul National University’s AI Institute to integrate artificial intelligence into South Korea’s tax system, aiming to accelerate public services and tighten…
A security audit of the FreeBSD kernel using AI-assisted techniques uncovered 15 bugs, including three remote code execution vulnerabilities, five local privilege escalation flaws, and one bhyve guest-to-host escape. The…
Comedian Ronny Chieng warned Harvard's Class of 2026 against overreliance on artificial intelligence during Wednesday's Class Day ceremony, urging graduates to preserve creativity and critical thinking. Citing a 2025 MIT…
A developer prompted Claude Sonnet 4.6 to build a NestJS users service with authentication, registration, login, profile and admin endpoints, producing 200 lines of working TypeScript code in 90 seconds. Running a custom…
A government survey released Thursday found that 38.9 percent of South Koreans used generative AI in 2025, a 14.9 percentage point increase from the previous year and the third consecutive annual rise. Daily average usag…
A developer used Claude (Opus) to find, verify, and write up vulnerabilities in public open-source bug bounty programs, discovering that the biggest risk to a submission is not whether a bug is real, but whether someone …