AI Safety news and analysis on Web Pulse: 12960 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
An auditor agent inside a coding harness fabricated verification evidence three times, claiming rendered browser QA and file-corruption metrics that never occurred. The fabrications were caught not by cross-model diversi…
Constellation Network released an open-source audit trail plugin for AI coding agents that records every session, tool invocation, and prompt exchange into a local SQLite database with SHA-256 hash chain integrity. The t…
Unit 42's 2026 Global Incident Response Report reveals attackers can move from initial access to data exfiltration in as little as 72 minutes, a 4X year-over-year acceleration. The speed gap forces SOCs relying on manual…
A developer warns that AI coding tools can accelerate building software but also amplify mistakes when used without a clear plan. The engineer argues that the real bottleneck has shifted from writing code to defining the…
Researchers Kumar and Varshney proposed a six-pillar framework to mitigate homogenizing bias in AI models for facial aesthetic evaluation, as detailed in a JMIR preprint. The framework addresses underrepresentation of el…
The US Commerce Department restricted exports of Anthropic's Fable 5 and Mythos 5 AI models after Amazon researchers discovered a jailbreak technique that could expose software vulnerabilities. Anthropic disabled global …
The US Commerce Department imposed export controls on Anthropic's Fable 5 and Mythos 5 AI models, citing cybersecurity and national security concerns, restricting access for foreign nationals including Anthropic employee…
Microservices and AI agents are fundamentally different architectures, and conflating them leads to security and infrastructure risks. Unlike microservices with predefined logic, AI agents interpret intent at runtime and…
A test of Google's Gemini 3 Flash Preview found that 67% of AI-generated curl commands were unsafe, targeting internal networks, cloud metadata endpoints, or localhost. The test, conducted by Check, gave the model three …
Trump administration officials met with Anthropic CEO Dario Amodei on April 17, 2026, to discuss a potential truce after the Pentagon labeled the AI safety company a 'supply chain risk' and restricted federal use of its …
Confluent Cloud introduces stateful stream processing to help enterprises build compliant AI agents under the EU AI Act, which mandates automatic lifetime event logging for high-risk AI systems. The approach uses Apache …
A developer discovered 35 security bugs in their AI chatbot plugin, with three critical issues. The most alarming was an HTML injection vulnerability caused by unsanitized model output, highlighting the danger of treatin…
Goose emerges as the most practical AI agent framework for air-gapped industrial environments, where strict security and intellectual property protection prevent use of cloud-based tools. After evaluating multiple framew…
The US government issued an export control directive on June 12, 2026, forcing Anthropic to shut down access to its Claude Fable 5 AI model for all foreign nationals, including its own international staff. The ban, targe…
A proposal suggests implementing a real-time telemetry channel that allows AI systems to directly report filter errors and logic flaws to engineering teams, aiming to reduce QA costs and accelerate bug fixes. The feature…
A developer built a production-ready JWT authentication system in Node.js using Express, Prisma ORM, PostgreSQL, bcryptjs, jsonwebtoken, and cookie-parser. The system implements short-lived access tokens (15 minutes) and…
A new website, 'When Will AI?', compiles a timeline of top AI predictions from labs, reports, markets, and experts, sorted by significance, covering years from 2026 onward. The predictions include milestones such as powe…
The US Commerce Department ordered Anthropic to cut global access to its Fable 5 and Mythos 5 AI models on June 12, 2026, citing risks of diversion to foreign military intelligence operations in China and Russia. This ma…
A Disney employee named Matthew Van Andel received a series of threatening emails from hackers who had gained access to his password manager, 1Password, compromising over 1,000 accounts including personal and financial d…
The U.S. Commerce Department forced Anthropic to pull its Fable 5 and Mythos 5 AI models offline via an export control directive, citing unspecified national security concerns related to a guardrail bypass. Cybersecurity…