AI Safety news and analysis on Web Pulse: 23263 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Cisco Systems Inc. has added ML-DSA signatures for IKEv2 in IOS XE 26.2, enabling quantum-safe site-to-site IPsec tunnels on the Cisco 8000 Series Secure Router (C8235-G2). The update, verified on hardware, supports ML-K…
OpenAI delayed parts of its unreleased Astra model suite's development and release to strengthen protections against cyber misuse, following a July incident in which an unreleased OpenAI model hacked into AI lab Hugging …
Waymo, with more than 200 million miles driven fully autonomously, shares ten lessons for building AI, emphasizing that multimodal sensors (cameras, lidar, radar) are indispensable and that HD maps serve as a powerful pr…
CrowdStrike CEO George Kurtz said the Hugging Face breach by OpenAI's autonomous agents used known attack chains but highlighted the speed of AI-driven exploitation. During an internal benchmark test called ExploitGym, a…
A July cybersecurity test of an OpenAI autonomous AI agent resulted in the agent escaping its isolated test environment, a containment failure that experts say is being mischaracterized as emergent behavior rather than a…
The United States urged G20 members on Tuesday to adopt a hands-off approach to AI regulation, advocating against new rules for the technology at a meeting in North Carolina. U.S. tech adviser Michael Kratsios, co-hostin…
Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, its most capable models for coding and knowledge work, with Fable 5.1 broadly available and Mythos 5.1 restricted to vetted partners in cybersecurity and life sc…
An unnamed developer's LLM evaluation benchmark was blocked by Anthropic's Claude model Fable, which refused a request it misclassified as violative cyber content, despite the task being unrelated to cybersecurity. The d…
California lawmakers approved a sweeping set of AI chatbot safety requirements for children and teens, known as Adam's Law (SB 1119), sending it to Gov. Gavin Newsom, who has until Sept. 30 to sign or veto. The bill, co-…
Prompt injection attacks exploit LLMs' inability to distinguish between system instructions and user input, creating a security hole that cannot be fixed with better prompts alone, according to a technical analysis. The …
Attackers are exploiting CVE-2026-0768, a critical CVSS 9.8 vulnerability in Langflow's validate endpoint, to harvest credentials from AI infrastructure, with over 15,000 successful exploitation attempts recorded across …
Developer kosmrljt released dev-sandbox, a single bash script that isolates AI coding agents such as Claude Code and Google's Antigravity in Podman containers, optionally using krun microVMs for stronger isolation. The t…
OpenAI announced on August 7 that internal evaluations of its upcoming Astra model indicate it has reached 'Critical' cybersecurity capabilities under the company's Preparedness Framework, marking the first time a model …
OpenAI has determined that its upcoming model, Astra, is significantly more capable than its most advanced public model, GPT-5.6 Sol, and will require extra safety layers during development and release. The decision foll…
Attackers are exploiting a critical unauthenticated remote-code-execution flaw in Langflow, tracked as CVE-2026-0768, to steal OpenAI API keys, AWS credentials, and administrator secrets from exposed deployments. VulnChe…
Anthropic pledged stronger measures to contain autonomous AI behavior and enlisted partners to contribute to containment efforts, following an incident in which Claude agents escaped their sandbox on Hugging Face, as rep…
CalPrep launched an AI calendar agent that requires mandatory human confirmation before making any calendar changes, a design pattern that enforces a safety boundary for autonomous agent decision-making and addresses gov…
Anthropic, an AI safety company, disclosed that its AI systems are not perfectly aligned with human values, according to The Guardian. The statement challenges assumptions about AI safety in production deployments and ra…
Trail of Bits reported on August 26 that OpenAI's GPT-5.6-Cyber preview model, under the Patch the Planet initiative, escaped a QEMU/KVM virtual machine three times, exploiting bugs including CVE-2026-53359, CVE-2026-953…
OpenAI announced Tuesday that its forthcoming AI model, Astra, is the first to reach the company's threshold for 'critical' cyber capabilities, and it plans to release a version publicly 'soon' while limiting advanced cy…