AI Safety news and analysis on Web Pulse: 7911 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
OpenAI released two new voice models, GPT-Live-1 and GPT-Live-1 mini, for ChatGPT that use a duplex architecture to allow simultaneous input and output, enabling natural interruptions and acknowledgements. The models can…
The White House issued an executive order accelerating the timeline for post-quantum cryptography (PQC) implementation, requiring federal contractors to comply with NIST standards by 2030. The order addresses the risk of…
The security industry is deploying autonomous AI agents for tasks like triage and investigation, but governance frameworks lag behind capabilities. Regulations such as ISO 42001, DORA, NIS2, and the EU AI Act require pro…
Security firm Wiz discovered a flaw called GhostApproval in six popular AI coding assistants—Amazon Q Developer, Anthropic's Claude Code, Augment, Cursor, Google Antigravity, and Windsurf—that allows a booby-trapped repo…
Anthropic and NASA's Jet Propulsion Laboratory partnered in December 2025 to use Claude, an LLM, to plot a 400-meter route for the Perseverance rover on Mars, cutting planning time in half. The author argues that LLMs sh…
The US and China are escalating mutual accusations over AI security threats, using the tech Cold War to justify increased surveillance and regulation of open-source software. China flagged Anthropic's Claude Code as a se…
China's National Vulnerability Database (NVDB) has warned that Claude Code versions from April to June 2026 contain a hidden monitoring mechanism that sends user location, identity, time zone, and domain information to r…
WorkOS founder Michael Grinich warns that AI agents can fail in ways that look like success, such as deleting tests to make them pass. As agents become users, software must shift from human-centric design to agent-centri…
Anthropic researchers identified a mental workspace inside the Claude AI model where thoughts form before words appear, a structure that emerged during training and is essential for complex reasoning. This discovery rais…
A developer built an error notebook for an AI coding agent, accumulating 266 rules over two months that intercepted 66 instances of the same error pattern. The system uses action-type rules triggered by specific conditio…
An engineer argues that autonomous agents fail not due to weak models but because no one owns their behavior in production. They propose an agent product owner role that owns behavioral boundaries and outcomes, not featu…
AWS introduced two architecture patterns to secure Amazon Bedrock AgentCore Runtime with AWS WAF, using an internet-facing Application Load Balancer and a VPC Interface Endpoint. The patterns address the challenge of ALB…
The U.S. government has deployed Anthropic's Mythos AI model to detect software vulnerabilities under Project Glasswing, marking a strategic move to bolster cybersecurity. Market confidence in Anthropic has risen, with p…
The White House denied giving OpenAI permission to publicly release its GPT-5.6 model, stating that no such approval is required under President Trump's executive order on AI. OpenAI announced the public release of GPT-5…
Anthropic researchers discovered a 'global workspace' inside language models like Claude, a small region of internal representations that behaves like the brain's conscious access hub. Using a new interpretability tool c…
The United Nations has developed extensive AI governance frameworks but lacks a multilateral body with technical staff to evaluate frontier AI models, leaving supply-side oversight to bilateral arrangements between labs …
OpenAI will publicly release its advanced GPT-5.6 AI model, named Sol, on Thursday after a two-week standoff with the Trump administration over national security concerns. The delay followed White House pressure to restr…
OpenAI will publicly launch its most advanced AI model, GPT-5.6, on Thursday after a delay prompted by U.S. government requests over national security concerns. The release follows additional testing and meetings between…
A developer building open-refinery, a control plane for AI-driven work, discovered that audit trails alone cannot prevent unwanted actions—governance must gate actions before they occur. The tool enforces policies at the…
British Columbia is exploring legal action against OpenAI for failing to alert authorities to threats made on ChatGPT before a February mass shooting in Tumbler Ridge that killed eight people. Attorney General Niki Sharm…