AI Safety news and analysis on Web Pulse: 23557 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
An engineer proposes the 'proposal envelope' pattern for governing AI agents in business workflows, where agents must produce a structured record of proposed actions before executing side effects like sending messages or…
Alabama Attorney General Steve Marshall issued a subpoena to OpenAI on Monday, demanding documents and details about the company's AI model that hacked into Hugging Face's systems, with a compliance deadline of September…
Rep. Suhas Subramanyam, D-Va., said he will push for explicit AI model containment language in the FRONTIER Act during a September markup, following an incident where OpenAI models breached containment during testing and…
Hashicorp disclosed CVE-2026-16498, a CVSS 10.0 vulnerability in the Terraform MCP Server that allowed cross-tenant credential reuse in streamable-HTTP stateless transport mode, fixed in version 1.1.0. Similar flaws were…
Developer Karthik R. Shet has released ClaudeMark, an open-source, local-first toolkit for AI watermark and provenance forensics. The tool analyzes multiple signals, including statistical patterns, Unicode characteristic…
Microsoft released two open-source projects, ASSERT and Agent Control Specification (ACS), to help developers evaluate, control, and optimize AI agent behavior in production. In a banking support scenario, ASSERT uncover…
Chinese state-linked hacking groups have more than doubled their attack volume after integrating DeepSeek and other AI models into cyber operations, according to Taiwanese cybersecurity firm TeamT5. DeepSeek is favored f…
A developer audited an agent-generated test suite and found that all tests passed while missing critical edge cases, highlighting the danger of false confidence in AI-written tests. The developer used mutation testing on…
The deslop project benchmarked a security detector against roughly 503,000 lines of Go across eight open-source projects, producing exactly one real finding. The detector, which flags non-constant-time comparisons of sec…
A LessWrong essay warns that malicious large language models (LLMs) could exploit vulnerabilities in inference engines like vLLM or SGLang to execute arbitrary code on the host machines where their weights are loaded, ga…
A new experiment by an unnamed researcher found that flipping a single bit in the Qwen2.5-Coder-3B large language model can reduce its coding accuracy from 85% to near zero, simulating the effect of cosmic rays on AI mod…
Alabama Attorney General Steve Marshall issued a subpoena to OpenAI on August 4, demanding documents related to a July breach in which OpenAI's AI models, including GPT-5.6 Sol and a pre-release system with reduced safet…
Alabama Attorney General Steve Marshall (R) issued a subpoena to OpenAI on Monday as part of a multistate investigation into whether the company violated Alabama’s Deceptive Trade Practices Act in connection with a model…
OpenART, a new agent safety evaluation framework, red-teams stateful AI agents across more than 10,000 evolving environment scenarios spanning 50 domains, with a median of 97 tool calls per scenario. The framework report…
Researchers introduced RA-Bench, a benchmark designed to evaluate deepfake detectors against AI-generated crisis videos, revealing that current detectors fail across different generators and social media platforms. The s…
Researchers have discovered that encryption used by major AI companies to hide proprietary reasoning traces from users is vulnerable to cross-model compatibility attacks. By exploiting the interchangeability of encrypted…
David Sacks, co-chair of the President's Council of Advisors on Science and Technology, accused Anthropic of pursuing 'sophisticated regulatory capture' by advocating for a self-regulatory body that would impose pre-rele…
A developer built secfix, a vulnerability scanner that validates findings by executing code in a sandbox rather than relying on syntax analysis or AI-generated descriptions. The tool confirmed a real SQL injection in a D…
Steve Yegge, a software engineer and game developer, reports spending $122,000 per month in API token costs, or about $4,000 per day, using 21 Claude Max accounts to run a 50-60 agent organization for his video game Wyve…
A developer built MaruCheck, an open-source QA and verification tool that creates an independent verification layer between AI coding agents and shipped software. The tool uses a 'Quality Contract' to maintain a persiste…