AI Safety news and analysis on Web Pulse: 9440 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
On June 12, the US government ordered Anthropic to disable Fable 5 globally within six hours, leaving developers without access and highlighting the risk of regulatory takedowns. The following day, GPT-5.6 Sol launched e…
GitHub Copilot's Terraform code suggestions often introduce severe security vulnerabilities, such as opening security groups to the world or enabling public access on RDS instances, because the suggestions are syntactica…
Chainguard CEO Dan Lorenc warns of a 'messy' summer for security teams as AI-powered bug-hunting models uncover thousands of previously hidden vulnerabilities in open source code. The Athena coalition, formed by Chaingua…
Chainguard launched Athena, a coalition of over two dozen companies using AI to find and fix open-source vulnerabilities before public disclosure. The group has processed more than 20,000 findings and produced over 2,000…
A new proposal outlines a runtime-enforced checklist mechanism to prevent AI agents from guessing or filling gaps in user instructions, aiming to reduce errors and accidents as voice-controlled agents and APIs expand. Th…
Anthropic received U.S. government authorization to allow a small group of American cybersecurity firms to access its powerful AI model Mythos 5, which had been blocked due to national security concerns. The company will…
Amazon Ring's AI-powered 'Familiar Faces' feature, which identifies and names regular visitors via facial recognition, has sparked backlash from neighbors and privacy advocates who never consented to biometric scanning. …
A new study measuring AI adherence to software architecture rules found that even advanced models like Opus ignore them 60% of the time, raising concerns about reliability in structured development environments.
OpenAI launched GPT-5.6 Sol, Terra, and Luna as a restricted preview limited to trusted partners at the request of the U.S. government, marking a shift toward government-mediated frontier AI releases. The flagship Sol mo…
A team of AI agents governed by five plain-text files runs three agents in production, handling strategy, architecture and coding, and reconnaissance. The governance system uses color-coded zones, a five-question preflig…
A developer built devops-mcp, a mode-based MCP server that lets AI assistants like Claude Desktop, Cursor, and Windsurf operate Linux servers with restricted permissions. The tool enforces a consent gate for state-changi…
More than 40 government officials from Central Asia, the Caucasus, and Asia-Pacific gathered in Tbilisi for a UN Public Service Forum workshop on balancing AI efficiency with public interest, discussing transparency, dat…
Students in East Asia are using AI-powered smart glasses to cheat on high-stakes exams, with recent incidents reported in South Korea and Taiwan. Educators and authorities are scrambling to update exam security measures …
A rapid sell-off in AI chip stocks this week exposed a roughly $270 billion speculation machine made up of leveraged positions, newly launched funds, and concentrated retail exposure, according to Bloomberg. The unwind, …
A vulnerability in Amazon Q Developer, tracked as CVE-2026-12957 with a CVSS score of 8.5, allowed attackers to steal AWS credentials by tricking developers into opening a malicious repository. The flaw, discovered by Wi…
Anthropic's Claude Mythos 5, its strongest cybersecurity model, is being redeployed to select US organizations that operate and defend critical infrastructure after receiving government approval. The company is working t…
OpenAI launched a limited preview of its GPT-5.6 series, including three variants—Sol, Terra, and Luna—to a small group of trusted partners at the request of the US government. The company plans a broader release in the …
Guyana President Dr. Mohamed Irfaan Ali urged Caribbean accountants to embrace artificial intelligence and proposed a regional cybersecurity training programme, warning that digital security is becoming an essential prof…
Verigate, a cryptographic trust infrastructure for AI agents, generates tamper-evident authorization receipts using Ed25519 signatures and Merkle trees to comply with EU AI Act Article 14. The system produces determinist…
A developer argues that production AI systems must wrap stochastic large language models inside deterministic software shells to ensure reliability. The approach uses constrained decoding, finite state machines, and form…