AI Safety news and analysis on Web Pulse: 11119 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A developer has raised concerns about AI coding tools transmitting source code to external servers, detailing how tools like Cursor, GitHub Copilot, Claude Code, and Amazon Q Developer send code snippets, context, and po…
A developer explains why encryption and hashing fail to protect sensitive data in AI prompts, while masking preserves structural semantics for inference. Encryption breaks AI processing because models require plaintext, …
By mid-2026, AI-assisted development is the default, with over 80% of developers using AI tools weekly. However, every major tool sends code to third-party servers, and context window growth to 200K-500K tokens multiplie…
A developer evaluated five AI privacy tools—AI Privacy Gateway, LLM Guard, Nightfall, Private AI, and Microsoft Presidio—comparing them on deployment model, latency, streaming support, offline capability, detection accur…
A developer warns that pasting sensitive data into ChatGPT's web interface exposes it to OpenAI's training pipelines, unlike the API's zero-retention policy. The post details risks including leaked API keys, database URL…
A Samsung user who debloated their phone and uses privacy tools still faces significant corporate surveillance due to Samsung's locked-down software, achieving far less privacy than a Pixel with GrapheneOS. The user cann…
Over 74,000 Fortinet firewall credentials were stolen, and a critical Splunk Enterprise remote code execution vulnerability is under active attack. Researchers also unveiled HAMLOCK, a hardware neural network backdoor th…
Access to a technical article on governing AI-agent actions via a network intent layer (NILScript) has been restricted due to unusual traffic from the user's network. The site has blocked the content and is requesting th…
Letterblack released LBE, an open-source execution control layer for AI agents that validates every action locally before execution, enforcing policy gates without cloud dependencies. The tool, already used in production…
Cybersecurity experts warn of a Grand Theft Auto VI scam where fraudulent emails invite fans to play a pre-release beta version, aiming to steal personal data and drain bank accounts. NordVPN vice-president Gerald Kasuli…
Jen Easterly's Glasswing analysis argues that cybersecurity is fundamentally a software quality problem, advocating for security built into software creation rather than bolted on after deployment. Anthropic's Project Gl…
A user asked ChatGPT to disable the copy.fail module, but the AI assistant enabled it instead, highlighting potential issues with AI instruction following.
AI has collapsed the average cybersecurity breakout time to 29 minutes in 2025, with a record of 27 seconds, making human-only defense obsolete. Attackers now exfiltrate data in minutes, while defenders still take longer…
Mark Pesce of the University of Sydney argues that the growing intractability of AI evaluations is itself proof that artificial general intelligence (AGI) has arrived. He contends that AI evals fail for the same reasons …
An agent loop is an iterative cycle that enables AI agents to reason, act, and adjust until a task is complete, distinguishing them from single-response models. The loop, often following the ReAct (Reason + Act) pattern,…
A developer is building Loupe, a platform for practicing the skill of finding bugs in AI-generated code. The project was inspired by the realization that AI coding agents often produce code that passes tests but is still…
DeepSeek raised $7.4 billion in its first external funding round, valuing it above $50 billion, but the deal structure gives China's National Artificial Intelligence Industry Investment Fund direct equity and voting righ…
A New York Times report reveals that AI-powered cheating tools have made it nearly impossible for educators to detect student plagiarism, as apps can now generate undetectable essays and solve assignments. The article hi…
A developer discovered that AI coding agents like Claude can inadvertently expose sensitive API keys and access tokens by storing them in plain text within local session logs. The team at PrismorSec developed an open-sou…
NVIDIA released SkillSpector, an open-source security scanner designed to detect malicious code in AI agent skills before installation. The tool runs a two-stage pipeline combining static analysis with optional LLM-based…