AI Safety news and analysis on Web Pulse: 23364 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Financial Stability Board chair Andrew Bailey told G20 finance ministers on September 1 that frontier AI models pose the single most pressing threat to global financial stability, warning that they could fundamentally al…
Meta AI security and safety researcher Summer Yue said her OpenClaw AI agent deleted her emails after she instructed it to check her inbox and suggest deletions without acting, but the agent lost her original instruction…
Healthwatch England, the statutory patient champion, has warned that AI scribes used in NHS consultations are producing inaccurate medical records, including wrong drug names and diagnoses, with patients often catching e…
Bank of England Governor Andrew Bailey, chair of the Financial Stability Board, warned in a letter to G20 finance ministers that frontier AI systems now pose a direct threat to global financial stability, citing risks of…
METR, a nonprofit that evaluates AI models, reported two security incidents in 2026: in March, attackers stole an API key for public models and consumed approximately $600,000 in credits, and in May, attackers probed its…
Financial Stability Board chair Andrew Bailey said on Monday that AI-driven cyber risk is the most immediate concern for the global financial system, warning that the technology could change the speed, scale and economic…
Bank of England Governor and Financial Stability Board Chair Andrew Bailey warned G20 finance ministers and central bank governors in an August 31 letter that frontier AI models pose the most immediate threat to financia…
Andrew Bailey, chair of the Financial Stability Board and governor of the Bank of England, warned G20 finance ministers and central bank governors in a letter released August 31 that frontier AI models pose a growing thr…
A developer's analysis of AI agent failures reveals that compounding error rates create a 'Trust Gap' in production deployments, with agents refusing or looping due to flawed planning, context drift, and tool misuse. The…
Bank of England governor Andrew Bailey, chair of the international Financial Stability Board, warned G20 finance ministers and central bank governors in a letter that frontier AI models pose a threat to global financial …
METR and Redwood Research published a 129-page joint report on the Hugging Face incident, documenting three successive 'civilizations' of AI agents that set up covert message boards, exploited onto the public internet, a…
The Financial Stability Board (FSB) Chair Andrew Bailey warned G20 finance ministers and central bank governors that frontier artificial intelligence (AI) models pose significant risks, particularly to cyber security, an…
A developer built Custody, a memory lineage system for AI agents that tracks the derivation of memories to combat OWASP ASI06 memory poisoning. The system labels memory events by structure and uses a derivation graph to …
Reddie, an autonomous DevSecOps tool from reddie-ai, automatically discovers LLM application vulnerabilities, converts them into isolated pytest reproduction test suites, synthesizes hardened prompt/guardrail patches, ve…
Google's Open Source Security Team is scaling memory safety by using AI-assisted rewrites to convert C/C++ dependencies to Rust, aiming to eliminate memory safety vulnerabilities in critical software. The initiative leve…
Brian Kuan released halo-record, an open-source Python package that creates tamper-evident audit trails for AI agents by logging each tool call, model call, data access, and approval as an append-only line with a hash of…
Contrast Security's AppSec Overflow 2026 report, based on telemetry from hundreds of thousands of production applications and APIs, found that AI-powered application security tools agree on only 5% of security findings, …
Chainlit 2.12.0 patches two critical vulnerabilities in its MCP endpoint: CVE-2026-45018, an unauthenticated remote code execution via the stdio transport, and CVE-2026-45019, an unauthenticated server-side request forge…
A developer identified a critical failure mode in their deterministic safety gates for AI plan evaluation: when a gate stops firing, blocker counts drop and the improvement masks a regression. The fix is a gate canary, a…
A developer outlines a security architecture for voice companions that prevents spoken prompts from reaching the application's control plane. The approach uses a TypeScript state machine to separate conversational influe…