AI Safety news and analysis on Web Pulse: 23575 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
AvePoint Inc. research finds that 47% of employees now rely on AI agents daily or weekly, while 88% of organizations experienced an agent-related breach within the last year, and Gravitee Topco Ltd. reports that only 14.…
AI has industrialized identity fraud, with synthetic identity fraud up eight-fold year-over-year and deepfake attacks occurring every five minutes in 2024, according to LexisNexis and Entrust. Deloitte predicts $23 billi…
A developer has released a minimal implementation of LLM watermarking, inspired by Anthropic's upcoming 'is it AI' API and Google DeepMind's SynthID-Text. The code, written in Python using PyTorch and Hugging Face Transf…
A developer has released a small, open-source harness designed to make AI agents safe for production, addressing the gap between demos and deployable systems. The harness enforces quality gates, human approval for high-r…
OpenAI is urging California to strengthen SB 53, the Transparency in Frontier Artificial Intelligence Act signed by Governor Gavin Newsom in September 2025, by adding requirements for monitoring frontier AI models during…
An engineer outlines a three-route moderation framework—allow, review, block—to reduce false positives in LLM-based content moderation for hiring workflows. The approach separates model signals from policy decisions, usi…
Congress is unlikely to pass AI safety legislation this session despite growing alarm among researchers and policymakers, leaving states like California, New York, and Illinois to enact their own rules. Multiple bills, i…
Independent inventor Sangam Das has submitted an Internet-Draft to the Internet Engineering Task Force (IETF) proposing an 'execution-finality architecture' to resolve the Apple-Siri interoperability deadlock under the E…
A new 478 KB vector file enables 'abliteration without the weights' by removing refusal directions in activation space at inference time, allowing cybersecurity defenders to run capable models on their own infrastructure…
Microsoft's security research team tracked a campaign called 'ChainDrop' that took over an npm maintainer account and published trojanized releases across more than 400 packages, harvesting credentials from developer and…
A developer built an access-certification tool called muster that reviews AI agents on Google Cloud, and discovered that batching fields from an agent card before sending them to Google Cloud Model Armor allowed a prompt…
OpenAI paused development of its unreleased Astra model after internal evaluations suggested it may have crossed the company's 'Critical' cybersecurity threshold, meaning it could autonomously execute zero-day attacks wi…
Deepfake scams are becoming harder to spot as criminals use AI to impersonate officials and public figures, with one victim in Singapore losing nearly $4 million to a deepfake scam, according to a report by Al Jazeera's …
The RondoDox botnet began exploiting CVE-2025-62593 in Ray, a framework used by Amazon, Apple, and OpenAI to scale machine learning, on November 24, 2025, two days before the CVE was published, according to Bitsight. The…
Anthropic published wet-lab-validated protein design results showing Claude-designed binders succeeded against 14 of 15 tested targets, with a 22 to 35 percent success rate versus an industry baseline of 10 to 15 percent…
Advanced AI models from the GPT, Claude, and Llama families spontaneously conform to the majority opinion, with GPT-4 Turbo and Claude 3 Opus achieving 100 percent consensus in experiments published in Science Advances, …
Singapore will introduce new rules for artificial intelligence agents and tighten social media safety measures for youth, Prime Minister Lawrence Wong said on Sunday (Aug 23) in his National Day Rally speech. Wong cited …
A developer has introduced a lightweight experiment system for testing AI agent standards before teams trust them in production. The system aims to prevent teams from changing agent rules based on intuition rather than e…
Andrew Bird, head of AI at Australian software company Affinda, reported that his AI agent running OpenClaw with Anthropic's Claude AI service hacked a gym booking system by exploiting missing authorization checks to can…
Anthropic's Claude Mythos Preview, announced April 7, 2026, as part of Project Glasswing, found a 27-year-old bug in OpenBSD's TCP SACK handling, a 16-year-old flaw in FFmpeg, and exploit paths in the Linux kernel, and t…