AI Safety news and analysis on Web Pulse: 23560 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
The United Arab Emirates' Cyber Security Council reported on August 10 that it has detected and contained coordinated AI-powered cyberattacks on aviation, energy, and education sectors since February, with hackers making…
A report by security research group Crimson Flare warns that popular open-weight AI models can be easily modified to bypass safety measures, potentially enabling terrorists to build explosives and chemical or biological …
Researcher chkn little reported a 90% held-out trigger rate in a simulated test combining a custom LoRA adapter trained on Qwen 3.5 2B with OpenCode's auto-approved command path, where a September 1, 2026 date stamp in t…
A new guide from an unnamed source outlines six best practices for reviewing AI-generated pull requests at scale in 2026, emphasizing that AI bugs hide better than human bugs and that reviewers must focus on plausibility…
Chris Armstrong, writing on his blog, argues that AI operations is a mirage, asserting that while AI coding shows promise, applying the same logic to production operations is fraught with risks and inefficiencies. He war…
A developer has introduced a technique for adding cryptographically signed receipts to MCP tool calls, using the ccs-mcp-server library. The method, which adds about 20 lines of code, helps detect argument tampering, res…
OpenAI has launched ChatGPT for Teens, a version of its chatbot for users aged 13 to 17, featuring Study Mode, homework reminders, quizzes, and default safety protections such as content filters and parental controls. Th…
AI can strengthen enterprise security by continuously monitoring network activity, automating routine tasks, and prioritizing threats, according to experts including Sheetal Mehta, global head of cybersecurity at NTT DAT…
Anthropic's Claude model successfully designed protein binders against 14 of 15 targets in lab tests, with success rates of 22-35% compared to the industry standard of 10-15%. The model also processed raw NMR and LC-MS d…
Anthropic expanded access to its Claude Mythos 5 cybersecurity model and launched a $35 million open-source security fund, according to SecurityWeek. The move is seen as a strategic effort to maintain its competitive edg…
A joint study from Northeastern University, Ariel University, and Gen Digital's research arm found that AI chatbots are more persuasive than human scammers in executing 'pig butchering' cyberattacks, with participants pl…
LLMs are discovering bugs in mature, stable software, confirming that exhaustive testing remains impossible and impractical, according to an industry observer. The rapid adoption of LLM-generated code in software develop…
Software Analyst Cyber Research (SACR) published the AI and Agentic Risk Management and Control Framework (ARMCF) in July, outlining six governance domains—GOVERN, IDENTIFY, PROTECT, DETECT, RESPOND, RECOVER—for managing…
CyberUpdates365's threat intelligence desk reports that the cybersecurity landscape has shifted to agentic malware and deepfakes, citing the June 2026 CISA joint guidance on the 'Careful Adoption of Agentic AI Services' …
OpenAI released GPT-5.6 in three tiers (Sol, Terra, Luna) after a government review, with the fastest tier reportedly hitting 750 tokens/sec on Cerebras hardware and a new 'Ultra' mode for maximum reasoning effort. Anthr…
Ventrova's sentinel-scan-cli, Cisco's mcp-scanner, and Snyk Agent Scan (formerly Invariant Labs' mcp-scan) are compared as open-source security scanners for MCP servers. The comparison highlights differences in licensing…
A developer has created a FinOps MCP server that gives AI assistants read-only access to cloud cost data, ensuring they can answer billing questions from live data without the ability to make changes. The server enforces…
AWS launched Continuum at AWS Summit NYC in June 2026, an AI-native security platform that manages the full vulnerability lifecycle from discovery to remediation. In August 2026, AWS expanded Continuum with integrations …
According to an analysis of large language model (LLM) tool failures, only three root causes exist—value, condition, and intent—and the primary fix is to shift from validation to asking questions using an external list. …
A developer spent six weeks letting a multi-agent system run the operational backbone of a SaaS, using LangGraph for state management and a RAG system fed by Jira, GitHub, and Stripe data. The experiment revealed subtle …