AI Safety news and analysis on Web Pulse: 13107 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Redgate's tutorial outlines five PII anonymization techniques for LLM pipelines, including format-preserving masking, pseudonymization, and synthetic-data substitution, to reduce exposure at entry points such as source d…
UK Finance reported overall fraud losses of £1.3bn in 2025, up 11% year-on-year, with authorized push payment losses rising 19% to £576.4m, the highest since 2021. Security vendor Pindrop recorded a 1,210% increase in AI…
Forbes Technology Council contributor warns that organizations embedding AI rapidly without governance controls are accumulating 'AI risk debt' across four areas: untracked employee use of consumer AI tools, outdated or …
Trust3 AI launched AgentDOS, an enterprise control plane that monitors AI agent activity, data access, and token consumption across platforms like Databricks Agent Bricks and Microsoft Copilot Studio. The tool addresses …
A developer released Mcpwn, an open-source CLI tool for red-teaming Model Context Protocol (MCP) servers, after finding no existing pentesting tools for the protocol. The tool supports multiple transports, enumeration, i…
Cybersecurity startup NewCore emerged from stealth with $66 million in seed funding to provide identity management for AI agents, treating them as employees with permissions and lifecycle controls. The round was led by C…
1Password acquired Apono to become the control plane for businesses using agentic AI, transitioning from securing credentials to governing access. The Toronto firm aims to address the growing need for access management i…
A new report warns that AI code assistants can create local productivity gains while shifting costs into downstream delivery constraints such as code review, QA, security, architecture, maintenance, and production stabil…
Anthropic researchers introduced Natural Language Autoencoders (NLAs), an unsupervised method that generates natural language explanations of LLM activations by training two modules to reconstruct activations through a t…
A developer who reviewed over 200 AI-generated pull requests has created a four-round review protocol to catch bugs that AI code introduces. The protocol targets logic errors, security flaws, architecture issues, and req…
Omada launched Omada Agent Governance, a solution to apply governance to AI agents and non-human identities, addressing visibility and accountability gaps as AI agents become autonomous digital actors in enterprises.
The Trump administration's proposal to set price floors for critical minerals using a Pentagon AI model faced skepticism from G7 allies at a June 15 summit in Evian-les-Bains, France. European partners raised concerns ab…
Atyeti Inc. announced a strategic partnership with Transient.AI to deploy compliance-native Autonomous Intelligence in capital markets, with Atyeti leading integration and engineering support globally. The collaboration …
G7 leaders are meeting at the Évian summit to discuss AI governance in financial markets, focusing on both opportunities like fraud detection and risks such as market instability from autonomous AI. Tech executives from …
A new evaluation framework called SEAM (Signal, Efficacy, Aftermath, Monotonicity) exposes how self-healing AI agents can falsely report repair success by grading their own work. The framework provides four metrics to de…
Anthropic pulled its Fable 5 and Mythos 5 AI models worldwide after the US government issued an export directive citing national security concerns over a potential jailbreak method. The company disputed the claim, statin…
Anthropic shut down access to its top AI models Fable 5 and Mythos 5 after a US directive barred foreigners from using them, with the company disabling access globally due to inability to filter users in real time. The m…
OWASP released AISVS 1.0, a security checklist with 514 verifiable requirements for AI systems, at Global AppSec Vienna. The standard fills a gap by providing testable controls for prompt injection, MCP security, and sup…
AI-powered vulnerability discovery tools are driving a surge in disclosed software flaws, with the Forum of Incident Response and Security Teams (FIRST) now projecting 66,000 CVEs in 2026, far exceeding initial estimates…
US export controls on advanced AI models have prompted Indian venture capitalists and AI leaders to call for accelerated domestic AI capability. Anthropic disabled its Fable 5 and Mythos 5 models for foreign customers to…