AI Safety news and analysis on Web Pulse: 12881 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
OpenAI researchers tested whether public chat data from WildChat can predict real-world AI misalignments, finding that deployment simulations using public conversations can estimate rates of undesirable model behavior, o…
KPMG, a global consulting firm with 250,000 employees, published an AI-generated report in October 2025 titled 'Total Experience: Redefining Excellence in the Age of Agentic AI.' An audit by GPTZero found that 40 of 45 c…
The US government pressured AI company Anthropic to restrict access to its latest models, 'Fable 5' and 'Mythos 5', to US citizens only, highlighting Europe's digital sovereignty challenges ahead of the G7 summit.
Tide's team of five engineers and designers built a verifiable authentication system that eliminates the need for a central credential store, removing a common breach vector. The team includes cryptography experts, a Key…
QuantaMind built a headless CLI tool to shift AI evaluation left into CI/CD pipelines, allowing teams to gate model merges based on agentic reliability thresholds. The tool integrates custom eval JSON collections into th…
An attacker hijacked a Mastra maintainer's account and republished 116 packages in the @mastra catalog over 27 minutes, adding a hidden dependency on the typosquat package easy-day-js. The malicious package disables TLS …
South Korea became the fourth country to form an AI security alliance with OpenAI, signing an MOU to collaborate on global AI safety evaluation frameworks and develop assessment standards reflecting Korean language and s…
The US Commerce Department has delayed adding Chinese AI developer DeepSeek, memory chipmaker CXMT, and over 100 other Chinese tech firms to the Entity List for nearly eight months, the longest gap in over a decade. The …
A former effective altruist explains how reading Ayn Rand's 'Atlas Shrugged' led him to leave the EA movement, arguing that EA bundles distinct philosophical ideals—rationalism, impactful agency, maximalism, and utilitar…
A researcher proposes that human brains minimize bias through extreme overparameterization and high-learning-rate training on small diverse datasets, while LLMs minimize variance. This 'catapulting' hypothesis could expl…
A developer at a Fortune 100 manufacturer describes the production guardrails for enterprise AI agents, emphasizing that the most critical controls are pre-existing infrastructure like IAM, network egress, and secrets ma…
A developer argues that AI coding tools should be governed with the same rigor as infrastructure, drawing parallels to DevOps practices like blast radius analysis, auditability, and measurement. The author notes that whi…
Hacker group SHADOWBYT3$ allegedly stole 859 MB of data from Nintendo's employee feedback platform TINYpulse on June 13, demanding $2 million. Leaked messages show employees expressed concern about the push for Microsoft…
G7 leaders at the Evian-les-Bains summit discussed granting select 'trusted partners' access to advanced US AI models, following the Trump administration's export control directive that prompted Anthropic to disable its …
G7 leaders concluded a three-day summit in Évian with discussions on AI security risks and social media regulation, joined by OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, and Mistral AI CEO Arthur Mensch. French Pr…
The StepSecurity Threat Intelligence Team has identified that multiple @mastra npm packages have been compromised. The security breach was disclosed in a GitHub issue on the mastra-ai/mastra repository, with the team det…
A developer at a company building create-microservices-app deliberately broke their own AI-built booking app to test whether automated contract checks catch plausible-but-wrong code. The experiment showed that a machine-…
Anthropic sent security researcher Nicholas Carlini to Washington to address US government concerns about AI safety after export control directives forced the company to temporarily suspend global access to its Fable 5 a…
Researchers found that pruned large language models can pass multiple-choice benchmarks but fail to answer the same questions in open generation, creating a 'benchmark illusion.' The study, using multilingual question an…
A LessWrong post proposes applying exploratory modeling, a decision-making methodology used by RAND, to AI governance challenges. The project would build computational models to stress-test policies across diverse plausi…