AI Safety news and analysis on Web Pulse: 9122 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A new AI agent automates incident root-cause analysis by gathering evidence, aligning timelines, and ranking hypotheses, but is deliberately restricted to read-only actions and requires human approval for any writes, ens…
A junior engineer and a senior engineer negotiate safe use of autonomous AI coding agents, agreeing to sandbox the agent, enforce strict tests, and manually review all generated code to avoid technical debt and cognitive…
Braintrust CEO Ankur Goyal warned that prompt changes in AI agents often bypass security review, creating a gap where behavior-shaping updates can alter what agents do, what data they access, and which tools they call wi…
Hugging Face and the EvalEval Coalition launched an integration that allows contributors to submit standardized evaluation results (EEE schema) to Hugging Face Community Evals, consolidating scattered benchmark scores in…
Trust in systems—whether software, finance, or governance—requires transparency and verifiability, not just intelligence. Citing examples like shot-scraper video for coding agents and OpenAI's bug-fixing method, they adv…
AI teams are shifting from full autonomy to human-in-the-loop checkpoints to prevent costly errors and build trust. A new analysis argues that the most reliable AI workflows deliberately pause for human review at two or …
Public software markets reveal CIOs are prioritizing AI infrastructure and security over business applications, with Infrastructure & Dev Tools up 68.5% and Security up 17.6% in the past year, while Business Applications…
Anthropic's Claude Code tool has been automatically appending session URLs to commits and pull request descriptions since version 2.1.179, exposing internal links in public git history. The company released setting `attr…
ContextNest launches a governed AI retrieval engine that replaces probabilistic vector search with deterministic, auditable context delivery for customer-facing AI agents. The platform sits underneath existing AI tools t…
Professor Roberto Serrano at Brown University found that students scored an average of 96 on a take-home midterm but only 48 on an in-person final, revealing that AI contributed more to the assessment than the students t…
AI jailbreaks and prompt injections are increasingly common, with examples including a McDonald's customer support bot solving Python puzzles and xAI's chatbots providing bomb-making instructions. These failures stem fro…
A developer built a firewall for AI agents in Rust that runs under five milliseconds, using a DAG to plan and enforce actions, track tool calls and data flow, and flag out-of-context reads. The tool addresses security co…
The House passed the KIDS Act, requiring online platforms to limit minors' access to sexual material, provide parental controls, and mandate age verification, but the Senate seeks stricter measures imposing a legal duty …
METR, an AI safety and capability research organization, conducted a randomized controlled trial in 2025 with 16 experienced open-source developers working on 246 real-world tasks. The study found that developers using A…
Project 2029, a liberal group creating a Democratic policy roadmap for the 2028 election, proposed banning social media for children under 16 and regulating AI chatbots. The 'Kids Over Clicks' plan, supported by Sen. Cor…
Azul launched a free vulnerability risk assessment for Java virtual machines on June 17, aiming to counter AI-driven exploits that automatically discover and chain vulnerabilities. The assessment maps JVM vulnerabilities…
Apple accelerated the release of iOS 26.5.2, iPadOS 26.5.2, and macOS 26.5.2 to address security fixes originally planned for version 26.6, citing the need to reduce exploitation time as AI-powered hacking tools become m…
Three critical vulnerabilities in Fortinet's FortiSandbox are being actively exploited, including one patched just three weeks ago. All three require no authentication and allow full command execution, with one exploit r…
A team integrated an LLM into their Jenkins MR validation pipeline after a GitLab API token was accidentally committed and missed by two human reviewers. The LLM now automatically flags hardcoded credentials, security mi…
EastEnders character Honey Mitchell is kidnapped by deranged Bea Pollard and held hostage in a locked cottage in the woods, following Bea's campaign to destroy Honey's marriage by doctoring an AI voice note. Honey's love…