AI Safety news and analysis on Web Pulse: 10583 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Varonis Threat Labs disclosed SearchLeak, a three-stage vulnerability chain in Microsoft 365 Copilot Enterprise that can exfiltrate emails, MFA codes, calendar items, SharePoint and OneDrive files via a single click. Mic…
A developer warns that AI coding agents like Claude Code, Cursor, and Copilot generate tech debt by confidently guessing dependencies from frozen training data, unable to consider internal libraries, package abandonment,…
Maturana, a new open-source AI agent platform, launches with hardware-isolated microVMs for each agent, zero-trust security via encrypted secrets and egress control, and a shared knowledge graph. Built on Codex and a mod…
Researchers at an undisclosed institution analyzed Gemma-2-2b's internal circuits and found that the LLM detects vulnerable code by relying on safety detectors that recognize safe coding patterns, rather than directly id…
Dr. Helmut Love launched AI-for-Good Georgia on Pride Day, a nonpartisan initiative to align artificial intelligence with humanity's highest moral and spiritual values. The initiative aims to bring together business lead…
A developer built EigenCV, a zero-trust resume pipeline that uses Infrastructure-as-Code to generate ATS-optimized LaTeX resumes without allowing AI to hallucinate skills. The system treats career history as an immutable…
Medical AI systems could compromise patient privacy through membership inference attacks, allowing hackers to determine if a specific patient's data was used to train a model, potentially revealing sensitive medical info…
Meta paused its Model Compatibility Initiative (MCI) employee-tracking program after an internal data leak exposed collected data to more employees than intended. The program, which tracked mouse movements, clicks, keyst…
A user expresses concern about medical privacy, citing AI scribes in healthcare and the risk of smartphone voice assistants recording conversations without consent. The user argues that banning phones and watches from me…
A user on Privacy Guides forum seeks GUI encryption tools for Linux that avoid AI-generated code and root privileges, citing concerns over Picocrypt-NG's use of 'vibecoded' AI code.
A developer with 13 years of experience describes how AI coding tools initially made them feel behind, but after embracing the technology, they built Ghostables to address client data security concerns. The developer als…
A developer found that different AI agents classify the same security flaw differently, highlighting inconsistent vulnerability assessment standards across AI systems. The experiment revealed that agents from major provi…
Coval, a San Francisco voice AI evaluation startup founded by former Waymo engineer Brooke Hopkins, raised a $28 million Series A led by Norwest to test voice AI systems with the same simulation-heavy approach used for a…
Leading AI researchers, including Yoshua Bengio and Geoffrey Hinton, estimate at least a 10% chance of human extinction from advanced AI, yet global response remains insufficient. Critics dismiss these as subjective prob…
A long-time participant in AI safety, Rationalist, and Effective Altruist communities reflects on the tendency to opine about these massive communities as a whole, despite only knowing small corners of them. The author d…
The e/acc (effective accelerationist) movement, often portrayed as a counterpoint to AI safety, lacks a coherent ideology, significant membership, or credible counterarguments to AI risk, according to a critical analysis…
Anthropic accused Alibaba's Qwen lab of using nearly 25,000 fake accounts to conduct 29 million exchanges with its Claude AI model, the largest distillation campaign yet. The accusation, shared with US senators and White…
A researcher built a bidirectional command-and-control channel using Bitwarden's icon proxy, embedding commands in PNG metadata and exfiltrating data via DNS to trusted Azure infrastructure. The attack exploits a confuse…
A new architecture using Azure Bot Service, Azure Functions, and Azure Service Bus enables secure Telegram connectivity for AI agents without exposing local machines to the internet. The approach eliminates the need for …
A user on a privacy forum is seeking free and open-source data poisoning tools for Linux or Android to protect their original photos from being scraped by AI systems. The user wants to disrupt AI training by making their…