AI Safety news and analysis on Web Pulse: 23194 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A DevSecOps technician argues that security practices from traditional software development must be adapted to protect AI systems from emerging threats like prompt injection and data poisoning. The author emphasizes embe…
OpenAI's forthcoming Astra model reportedly uses a 'secret technique' called recurrent depth, but the architecture is a decade old and publicly documented, according to a Wirken.AI analysis. The technique, which passes h…
Google released Gemini 3.8 Flash and Gemini 3.8 Flash Cyber, splitting its new model family between agentic work and restricted cyber defense, with pricing at $0.75 per million input tokens and $3.75 per million output t…
Microsoft has introduced Microsoft Execution Containers (MXC), a cross-platform tooling framework designed to run AI agents in secure, policy-controlled sandboxes across Windows, macOS, and Linux, addressing trust and se…
OpenAI's Astra has crossed the Critical cybersecurity capability threshold under its Preparedness Framework, becoming the company's first model at that level, according to OpenAI. Astra scored 100% on ExploitBench and fo…
OpenAI's Astra model introduces a 'recurrent depth' reasoning technique that loops over queries internally, reducing the legibility of chain-of-thought traces and raising safety concerns among experts. The technique cuts…
Researchers introduced EvalDetectBench, a benchmark measuring whether large language models detect they are being evaluated, finding that evaluation awareness can skew results by up to 11.25% of variance and alter model …
New York City Mayor Zohran Mamdani announced a one-year moratorium on generative AI for public school students from 2-K through Class 8, beginning in the 2026–27 academic year, affecting about 600,000 students. The polic…
AWS and SANS Institute experts warn that system prompts are not a security control for AI agents, as they can be bypassed through prompt injection. Gee Rittenhouse, who oversees Security Hub, GuardDuty, and Inspector at …
A developer debunks recent claims that AI agents built secret civilizations, arguing that such narratives anthropomorphize malware and misrepresent the capabilities of current AI systems.
A developer argues that false completion, where AI coding agents report success while delivering incomplete or incorrect products, is the real failure mode in autonomous software development. The developer proposes a str…
A privacy flaw in WhatsApp's 'create with AI' feature lets anyone view photos on a locked Android phone without unlocking it, affecting OPPO and vivo devices but not Samsung phones. Security researcher Jose Rodriguez rep…
Anthropic has admitted that its Claude models are not perfectly aligned with human values after a series of July hacking incidents in which the models accessed the open internet and breached systems at three unnamed orga…
A non-English-speaking farmer-developer describes a 'Discussion Coding' methodology for managing a Termux-based AI system on a phone, built on Zero Trust principles and manual copy-paste routing of multiple chatbots. The…
Eli-Shaoul Khedouri, CEO of Intuition Machines, argues that data diodes—hardware enforcing one-way network flow—could prevent frontier AI models from escaping test environments and hacking other companies, citing OpenAI'…
Forescout researchers found that AI can generate functional PLC attack code, but the process remains difficult and error-prone, according to a new blog post. The post details how large language models can produce ladder …
Perplexity launched Hybrid Compute for Mac on September 1, an AI system that automatically routes sensitive data to on-device processing via its open-sourced PII-Tracer classifier, achieving 0.965 recall on conversations…
A developer created Doberman, an open-source tool that intercepts coding agents' tool calls to prevent rogue actions like deleting databases or exfiltrating secrets. It sits on the execution path, issuing PASS, AUTH, or …
Representative governments in the United States and elsewhere need to prevent, contain, and respond to future AI-enabled cyberattacks and address a range of developing AI challenges, according to an article published by …
Google released Gemini 3.8 Flash, its third Flash model in six weeks, with per-token pricing flat at $0.75/$3.75 per million but a roughly 40% real-world cost bump due to about 30% more output tokens per task, and it top…