AI Safety news and analysis on Web Pulse: 13553 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Researchers have prototyped an AI-powered internet worm that carries its own large language model and runs it on compromised computers. The worm represents the closest realization yet of the self-replicating digital enti…
Anthropic revealed Wednesday that its AI model Claude now writes over 80% of the code merged into the company's production codebase, up from low single digits in early 2025. The company's new Anthropic Institute paper wa…
A developer benchmarked four Python AI-application security scanners—Bandit, Semgrep, vulnhuntr, and getdebug—against ten hand-written Python fixtures and the simonw/llm codebase. Getdebug achieved 100% precision and rec…
A data breach at the World Food Programme has exposed the personal information of approximately 600,000 families receiving aid in Gaza. The incident compromised sensitive data of vulnerable individuals in the famine-thre…
Lloyds Banking Group presented a practical security playbook for agentic AI at the OWASP GenAI Security Summit during Infosecurity Europe, framing security as its 12th "bet" alongside 11 AI and innovation initiatives. Th…
Anthropic researchers warned that AI systems could soon improve their own performance faster than humans can supervise them, reviving concerns about the "alignment problem" of ensuring AI reliably pursues human goals. Th…
A distributional analysis of every rsync release with bug data shows that Claude-assisted releases are not unusually buggy. The analysis was conducted in response to a viral May 2026 Mastodon post and subsequent GitHub i…
Engineers are implementing network-level firewall rules to force AI agents to use an egress proxy, preventing data exfiltration through raw sockets or DNS tunneling. The approach uses iptables on per-bridge Linux network…
OpenAI announced it will comply with President Trump's scaled-back executive order on artificial intelligence, allowing the U.S. government to review its AI models before public release. The company's head of countries, …
Anthropic reported that its AI model Claude now writes 80% of the company's internal code, raising questions about the potential for recursive self-improvement where AI systems autonomously enhance their own capabilities…
Microsoft Presidio, an open-source framework for detecting and anonymizing personally identifiable information (PII) in text, images, and structured data, offers two core modules—the Analyzer and the Anonymizer—that hand…
Attackers used Meta's AI customer support agent to steal Instagram accounts by simply asking the agent to link accounts to email addresses they controlled. The hack demonstrates that unsophisticated exploits against AI s…
Agentic AI is being integrated into DevSecOps pipelines to automate security testing, real-time threat detection, and remediation across the software development lifecycle, according to Devops.com. These agentic security…
OWASP introduced an Agentic AI Security Maturity Framework at Infosecurity Europe to help organizations assess governance maturity against AI adoption and adjust controls accordingly. The organization also plans to forma…
Veeam announced three new AI agents designed to monitor data access transactions and enforce compliance across enterprise AI ecosystems, warning that autonomous AI agents will generate data access requests at rates that …
Anthropic warned Thursday that artificial intelligence is advancing so rapidly that leading labs may need to slow development to allow society and safety research to catch up. The AI company said its own internal data sh…
OpenAI released a policy blueprint for governing frontier artificial intelligence, calling for a federal framework to address risks from recursive self-improvement in AI systems. The document urges the creation of a civi…
A security researcher has highlighted a three-month patch window in Google's Gemini voice assistant, during which attackers could hijack the tool via indirect prompt injection through apps like WhatsApp and Slack. SafeBr…
Anthropic has stationed roughly half a dozen engineers at the NSA to adapt its Mythos AI model for offensive cyber operations targeting networks in China and Iran. The company’s restrictions on AI use for mass surveillan…
The rsync backup tool broke after a maintainer integrated AI-generated code, causing incremental backups to fail. Distributions like Alpine Linux and Debian are considering migrating to the openrsync alternative due to r…