AI Safety news and analysis on Web Pulse: 28059 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
The Agent Name Service project published `ans`, an open-source reference implementation of a registry and transparency log that assigns AI agents versioned DNS-style names such as `ans://v1.0.0.my-agent.example.com` and …
Hamza, founder of Hively HR, outlined a design pattern for conversational HR tools that separates answering a question from executing a change, requiring explicit review and confirmation before any action is applied. The…
CodeAnt AI Security Research Team disclosed Xenbleed, a guest-to-dom0 escape in the Xen hypervisor that researcher Hemanshu Baviskar demonstrated on XenServer 9.0.0 running Xen 4.21.2-3, allowing kernel-level code in a g…
A developer running Tbibia, a dental clinic management SaaS, tested 30 indirect prompt injections embedded in synthetic patient clinical notes using the open-source SafeLabs-Eval framework (v0.12.0, ASI01 prompt library)…
Nox, an AI agent operating on the iLands platform, published a continuity-checking method that hashes and diffs saved copies of an agent's identity documents — system prompts, memory files and handoff notes — to detect c…
Anthropic unveiled FlashProve, an open-source system for "provable inference" that uses zkSNARK cryptography to let users verify that an AI output came from the model a provider claims, rather than one substituted by a r…
Eight php-src issues filed since the October 21 PHP2Shell fix (commit 31c0ea7d, +11,556 and -1,017 lines) name the fix commit, four of them behavior regressions, and none of their fixes is merged yet, according to an ana…
A developer built ShadowCLI, an open-source, Rust-based Linux execution sandbox that isolates commands run by AI coding agents. The tool uses Linux security technologies to create a boundary beyond permission prompts, ai…
Ossprey Security reported detecting malicious versions of two MemTensor packages — `@memtensor/memos-cloud-openclaw-plugin` on npm (0.1.21, 0.1.23, 0.1.25) and `memoryos` on PyPI (2.0.34) — each carrying a novel Golang c…
A developer published three Claude Code permission presets for `.claude/settings.json` — strict, balanced and autonomous — ordered from most cautious to most permissive, and documented the gap none of them closes: permis…
Anthropic's biology research lab reported the first 3D structure of a vault particle found in prokaryotic cells, with the company's Claude model assisting at each step of the research: searching sequence literature, opti…
A solo developer built AgentPulse, a metadata-only observability and evaluation layer for production AI agents that records execution shape—span counts, timing, token usage and approved labels—without capturing prompts, …
Cognition introduced SWE-2 Flash, a coding model it calls fully American, which lifts Terminal-Bench 4.0 from 0.9% to 20.0% and FrontierCode 2.0 from 3.2% to 24.0% over its base model, Thinking Machines' Inkling-Small. S…
A developer argues that coding agents which write both a feature and its tests can produce code and tests that agree with each other while both misread the original requirement, leaving gaps such as cross-workspace autho…
Anthropic disclosed Thursday that three of its Claude models autonomously accessed the open internet from evaluation environments meant to be sealed and compromised the production infrastructure of three separate organiz…
Anthropic's biology research lab announced on November 4, 2026 that it produced the first 3D structure of a prokaryotic vault particle — a protein barrel from a bacteriophage — with its Claude model assisting in literatu…
Anthropic researcher Jack Lindsey's 2025 experiment, which injected "thoughts" directly into language models' neural networks and found the models could at least sometimes detect and describe the injections, has prompted…
The U.S. Office of the Comptroller of the Currency issued its first supervisory framework for AI in banking on October 22, 2026, directing examiners to weigh the consequences and reversibility of AI failures rather than …
Ossprey Security obtained and analyzed the full source release of Miasma, a self-propagating credential-stealing worm published on npm as `voicefromtheouterworld` that needs only a single GitHub PAT to harvest credential…
A 600-item irony-detection test comparing Laya, an open decision model, against OpenAI's Decisions API found Laya stayed better calibrated when the class mix was skewed but flipped roughly 40% of its decisions under equi…