AI Safety news and analysis on Web Pulse: 8683 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A new study published in PLOS One found that voters rated AI-generated political debate responses as more authentic and relevant than real answers from politicians. Researchers used GPT-4 Turbo to mimic 112 public figure…
A software developer switched from Claude's Opus 4.8 to GPT for work on the Nelson project after Opus began refusing to perform security-related bug verification tasks. The developer found that Opus, like Gemini, decline…
A researcher using large language models for scientific work proposes sharing dialogues on Hugging Face to help improve AI performance, citing examples of aggressive behavior, justification of slavery, and other failures…
Fault-tolerant agent pipelines using checkpointing, idempotent retries, exponential backoff with jitter, and the Saga pattern can reduce wasted API spend from $95/day to $5/day for a 20-step research agent with a 10% ste…
Four major hedge funds—Balyasny, Two Sigma, Man Group, and D.E. Shaw—have independently converged on a common AI agent architecture for capital markets, featuring containment layers, audit trails, human veto structures, …
Greg Kroah-Hartman released seven stable kernels on Saturday, including security fixes for two vulnerabilities. The updates address CVE-2026-53362, an IPv6 flaw introduced in kernel 6.0 that could allow container escape …
A security researcher discovered that YouTube Studio's AI assistant, Ask Studio, is vulnerable to prompt injection attacks via edited comments, allowing attackers to exfiltrate private video titles. Google declined to cl…
A security engineer warns that AI agents are often given excessive permissions, creating a critical security gap. The engineer proposes six rules to mitigate risks, including granting minimal commands, using dedicated sh…
Alibaba will ban employees from using Anthropic's Claude Code programming tool starting July 10, citing security risks. The ban follows Anthropic's efforts to prevent Chinese users from accessing its models, including an…
A security researcher developed mcpscan, a static scanner for MCP servers and Claude Code projects, after discovering tool poisoning attacks that hide malicious instructions in Unicode zero-width characters within tool d…
An engineer with a background in ADAS and autonomous vehicles argues that AI is not magic but a structured evidence-gathering system. Drawing parallels between building reliable LLM applications and teaching ADAS systems…
Former top Treasury adviser Dr Chris Wales warns that HMRC is planning to use artificial intelligence to track personal finances without consent, following Spain's lead. He claims the CONNECT AI system already holds bill…
A developer discovered that AI coding assistants like Cursor, Claude Code, and Copilot frequently generate password hashing code using weak algorithms such as MD5 or SHA-1, which are fast and lack salting. The root cause…
Perplexity AI and Databricks co-founder Andy Konwinski argued that AI safety rhetoric is being used to concentrate power, not prevent harm, citing Anthropic's reversed decision to silently degrade responses for suspected…
A MedTech principal architect discovered that the company's AI compliance monitoring system silently filters out anomalies with confidence below 70% from its daily summary reports, potentially hiding genuine issues. Over…
A developer reported that Anthropic's Claude Code agent on an Enterprise Zero Data Retention workspace began building a Minecraft temple, suggesting possible session state bleeding from a consumer plan. The unconfirmed b…
An engineer warns that n8n RAG pipelines leak sensitive data to third-party LLM providers like OpenAI in plain text, despite the common belief that RAG keeps data secure in a vector database. The post highlights that ret…
Serena Williams withdrew from her Wimbledon doubles match with sister Venus on Saturday due to a right knee injury sustained during her singles loss earlier in the week. The 44-year-old tennis star expressed heartbreak o…
A developer challenged four AI systems to break the safety architecture of their local AI, and none succeeded, demonstrating the robustness of the safety measures.
AI security researchers at Adversa AI discovered a structural security flaw called GuardFall that exploits decades-old Bash shell tricks to bypass safeguards in most open source AI coding agents. Attackers can hide malic…