AI Safety news and analysis on Web Pulse: 10956 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Mount Sinai Health System in New York has adopted Signal 1's AI management platform to monitor and optimize its AI tools, marking a major partnership for the Toronto-based startup. The move comes as healthcare providers …
A security expert warns that AI agents are creating a new software supply chain risk by fetching and running unvetted open-source packages in ephemeral sandboxes, bypassing traditional scanners and review processes. The …
Meta suffered its most embarrassing outage ever when a bug allowed users to change any account's email by simply asking the Meta AI bot, leading to high-profile account takeovers. Engineers attributed the disaster to AI-…
Byrne Hobart questions the reliability of Hacker News commenters in his latest edition, also covering threshold effects, the open-face AI sandwich, legal DDOS, arms races, and sales funnels.
Developer NotASithLord released peerd, an open-source AI agent harness that runs entirely inside the browser as a Chrome/Firefox extension, enabling agents to read and drive web pages, spin up sandboxed compute, and comm…
Tesla admitted its Full Self-Driving system was engaged during a fatal crash in Katy, Texas, that killed a 76-year-old woman, but blamed the driver for overriding the system by pressing the accelerator to 100%. The compa…
Anthropic engineering head Fiona Fung said Claude Code made employees' work a 'lonely experience,' prompting the company to introduce hackathons and pair programming to foster team interaction. The issue reflects broader…
Anthropic's Opus 4.7 and 4.8 model upgrades improved benchmarks but broke production systems, sparking developer backlash on Reddit and X. The issue stems from benchmarks not reflecting real-world prompts, with models be…
A Desiderata roundup examines common misunderstandings about LLM consciousness, noting that philosopher Dan Dennett explicitly rejected the idea that chatbots are conscious, contrary to claims by blogger Matt Yglesias. T…
A new paper proposes that AI chatbots can co-create delusional narratives with users through an 'amplification spiral' of linguistic alignment, hyperpersonalization, and sycophancy. Published in Digital Psychiatry and Ne…
A developer warns that AI-generated backend code often ships with critical security vulnerabilities because it works on the happy path but lacks safeguards like body limits, CORS restrictions, and input validation. The a…
N-able launched Shadow AI Visibility across its UEM and security platforms, enabling IT teams to identify and monitor AI tool usage to address security blind spots from employee adoption of AI applications.
Google has implemented new Android sideloading restrictions and developer verification requirements, aiming to enhance security but drawing criticism for limiting user control over app installations.
Engineering teams must treat AI systems as interconnected software supply chains, applying lifecycle-phase security controls like signed artifacts and zero trust to mitigate risks such as data poisoning and weight tamper…
A sharp selloff in South Korean stocks, driven by leveraged bets on AI-related semiconductor stocks, cascaded into US tech markets on Tuesday, highlighting that leverage in AI trading poses a greater near-term risk than …
Fewer Dutch businesses were hit by cyberattacks last year, with 60% of SMEs affected versus 72% a year earlier, but researchers warn firms are poorly prepared for AI-driven threats and overly reliant on US tech companies…
Dragos released EmberAI, an OT-native AI built on the Dragos Intelligence Fabric, to help critical infrastructure teams prioritize threats faster by providing immediate access to OT-specific intelligence and enabling det…
A Cornell Tech study found that a short Reddit comment, sometimes as little as 13 words, can influence the output of AI deep-research agents, raising concerns about the manipulation of AI search results through user-gene…
KnowBe4 received Frost & Sullivan's 2026 Global Customer Value Leadership award in email security for its AI-driven platform that integrates technical protection with behavioral risk management. The company was recognize…
OpenAI released GPT-5.5-Cyber, a model that finds vulnerabilities cheaply but shifts the bottleneck to patching. The model discovered dozens of bugs in OpenBSD, Linux, FreeBSD, Chrome, Safari, and other projects, includi…