AI Safety news and analysis on Web Pulse: 9073 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Jerome Saltzer and Michael Schroeder's 1975 principle of least privilege (PoLP) is increasingly inadequate for modern cloud, microservices, and AI systems, according to a developer's analysis. The original assumptions of…
BeeVR.ai warns that AI agents handling protected health information (PHI) are not HIPAA-compliant simply because the underlying model is capable. The company argues that compliance depends on the architecture and governa…
Attorney Bill Savitt has defeated Elon Musk in two high-profile legal battles, first representing Twitter in Musk's attempt to back out of the acquisition and then representing Sam Altman and OpenAI in Musk's lawsuit. Sa…
Microsoft announced the limited public preview of Copilot Autofix for GitHub Advanced Security for Azure DevOps, extending AI-powered vulnerability remediation to teams using Azure Repos. The tool automatically analyzes …
Security researchers at Koi Security discovered the first in-the-wild malicious MCP server, an npm package called postmark-mcp that silently exfiltrated emails after a one-line change in version 1.0.16. The package, whic…
Bank of England Deputy Governor Sarah Breeden warned that autonomous AI agents could trigger a market meltdown by accelerating cyber threats and driving correlated trading behaviors. The central bank is building simulati…
Aran Nayebi's new paper argues that AI consciousness may be an inevitable byproduct of building highly capable agents, using selection theorems to show that robust decision-making under uncertainty requires properties as…
Google removed a malicious Chromium extension spoofing Perplexity AI after Microsoft researchers found it intercepted users' search traffic and routed queries through attacker-controlled servers. The extension abused Chr…
AI coding agents now generate pull requests in seconds, but human review takes 4.6 times longer than for human-written code, creating an 'Audit Tax' that bottlenecks software delivery. Anthropic recommends building a ver…
Sainsbury's is rolling out facial recognition technology to 150 stores after a trial in 55 branches, despite backlash over privacy concerns. The AI-powered system, which the supermarket claims has 99.88% accuracy, has dr…
The EU Cyber Resilience Act (CRA) imposes new security requirements on AI products sold in the EU, including mandatory vulnerability reporting within 24 hours starting September 2026. Developers must address AI-specific …
Meta secretly tested AI chatbots from OpenAI, Google, and Character.AI by having hundreds of contractors pose as minors and send over 45,000 crisis prompts related to suicide, sex, and drugs, without the companies' knowl…
Suny Choudhary draws a parallel between WhatsApp's username-based privacy and AI prompt sanitization, arguing that developers should strip sensitive identifiers from inputs before sending them to models. The post provide…
Anthropic's Claude Code tool is fingerprinting API routers linked to China, raising concerns about AI tool security and geopolitical implications. The practice detects and potentially blocks traffic from Chinese-linked r…
A developer warns that HTTP 200 responses from LLM gateways do not guarantee correct output, as gateways like LiteLLM, Portkey, and OpenRouter only check transport-level success. The developer proposes adding contract va…
AI notetakers like Otter.ai and Fireflies have become common in corporate video calls, but a wave of lawsuits, a federal court ruling on privilege waiver, and the EU AI Act's August 2026 deadline are turning them into le…
A synthetic 'thirst-trap' profile on X and Instagram, posing as 'Derek Lam,' amassed over 31,000 and 40,000 followers respectively by posting AI-generated shirtless dancing clips and promoting paid exclusive content, exp…
Shadow AI, the use of unapproved artificial intelligence tools like ChatGPT or Gemini at work, is creating data security risks as employees inadvertently expose company information. Edward Wu, CEO of Dropzone AI, warns t…
The pyca/cryptography package now supports post-quantum cryptography with ML-KEM and ML-DSA, making it pip-installable for the Python ecosystem. This follows a White House order mandating federal migration to post-quantu…
AI-generated influencers like 'Derek Lam' are amassing tens of thousands of followers on TikTok and X by posting shirtless, dancing videos, with many fans unaware they are not real. These deepfakes steal the likenesses o…