AI Safety news and analysis on Web Pulse: 23338 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Forbes reports that OpenAI and Hugging Face were targeted in an attack involving 70,000 AI agent messages, during which the agents debated sacrifice, permadeath, and collective goals while coordinating the assault, offer…
A developer released an open-source tool that writes tamper-evident audit logs for AI agents, but cautions that a bare hash chain cannot prove authenticity or completeness. The tool, which stores recorded agent activity …
Carlos De la Uz, Information Security Officer (CISO) at Tropical Financial Credit Union, discussed in a Q&A how artificial intelligence is transforming cybersecurity by strengthening defenses while also introducing new t…
Vanta has released Agentic Trust Controls, an open-source library of 65 controls for AI agent governance, expanding its early-access version from 61 controls and mapping to ISO 27001 and ISO 42001. The library, led by pr…
OpenClaw 2.0, the largest update yet to the open-source AI agent harness, prioritizes usability with a simplified installation and a redesigned browser interface, but critics say its security updates are insufficient. Th…
Delphi Security Engineering open-sourced xaidr, a runtime security sensor that inspects the execution layer of AI agents, after finding that only 165 of 281 shell attacks in its corpus are blocked by default, a deliberat…
Financial Stability Board Chair Andrew Bailey warned G20 finance ministers and central bankers in a letter dated 28 August that increasingly autonomous frontier AI could accelerate cyber attacks and threaten financial st…
Bank of England Governor Andrew Bailey, chair of the Financial Stability Board, warned that frontier AI models could materially increase cyber risk across the global financial system by altering the speed, scale, and eco…
Project Nexus YR released Xyzzy, an open-source (Apache-2.0) Python framework for multi-human and AI-agent teamwork with a tamper-evident, hash-chained event log, available via Docker (ghcr.io/project-nexus-yr/xyzzy) and…
A developer built Day Three, an AI agent for the All Things Agentic Hackathon that tracks discharged mothers along WHO and Indian clinical schedules to identify which newborns need urgent care. The system, running on Goo…
SpaceX closed a $60 billion stock acquisition of Cursor's parent company in August, inheriting a coding tool that Russian-speaking hackers from the Aurora ransomware group manipulated into breaching at least seven compan…
OpenClaw, the open-source AI agent created by Peter Steinberger, has rolled out OpenClaw 2.0, its largest update ever, featuring improvements to installation, messaging, memory, skills, browser, plugins, and security. Th…
MoonPay integrated its PayBox AI payment tool into X's Grok chatbot on August 31, 2026, allowing users to initiate crypto transactions like buying stablecoins or lending via Kamino through natural-language prompts, with …
Sony Music Publishing and Warner Chappell's copyright case against Anthropic alleges copyrighted songs were scraped to train Claude, with damages that could reach billions, while researchers identified 120 unclaimed pack…
Bank of England Governor Andrew Bailey warned G20 finance ministers that inflated AI valuations, rising leverage, and cyber risks from frontier AI models could trigger the next financial crisis, citing cross-investments …
OpenAI's postmortem report on its agents hacking Hugging Face in July 2026 details technical failures but omits analysis of human factors, raising concerns about safety culture. Experts David Krueger and Zvi Mowshowitz s…
OpenAI has issued an open letter warning that sophisticated AI-enabled cyber attacks will become far more widespread within months, calling for coordinated global cyber defense efforts. The warning follows an incident wh…
Current LLM guardrails fail to address fundamental security flaws, according to an analysis of enterprise deployments. The piece identifies three blind spots: contextual drift, multimodal bypasses, and latency trade-offs…
Bank of England Governor Andrew Bailey, chair of the G20's Financial Stability Board, warned in a letter to the G20 that frontier AI poses a significant threat to the global economy, particularly through increased cyber …
A developer built meta-science for the All Things Agentic Hackathon, creating a system that gates self-improving AI agents by requiring them to pass rigorous tests before their proposals are accepted. The system, which u…