SafetySpotify to Label AI‑Created Performer Profiles as “AI Persona” from Mid‑SeptemberSpotify will begin marking performer profiles that represent AI‑generated personas with an “AI Persona” label starting mid‑September, the company announced in a Wednesday statement.2 min read
SafetySubstack partners with Pangram to curb AI-written content — promise, limits and social risksSubstack has integrated Pangram’s AI-detection tools to give readers visibility into whether posts were written with AI, aiming to discourage low-effort machine-generated content.4 min read
SafetyTop AI Researchers in Las Vegas Urge Openness to Prevent Monopolies and Manage RisksAt the Ai4 conference in Las Vegas last week, Geoffrey Hinton, Fei-Fei Li and Andrew Ng debated how to balance openness and safety in AI development.4 min read
SafetyShared memory, research automation and new directions in Chinese large modelsRecent developments in AI show a shift from isolated model runs toward persistent shared memory, large-scale automated research, and agent-focused model designs — with repercussions for how models can improve over time.4 min read
SafetySophie Alpert: no lossless transformations of natural‑language textSophie Alpert published a short internal policy for engineers on acceptable use of AI for writing, arguing that every sentence in shared documents must reflect the author's own thinking.2 min read
SafetyResearchers recovered encrypted chain-of-thought from proprietary LLM APIs by replaying tracesResearchers demonstrated that Anthropic, OpenAI and Google models returned encrypted chain-of-thought blocks in API responses.3 min read
SafetyIndustry coalition proposes shared incident-reporting framework for AI agentsA coalition of more than 120 organizations led by the Open Secure AI Alliance has proposed a Shared AI Findings Exchange (SAFE) to standardize reporting of cyber incidents involving autonomous AI agents.3 min read
SafetySecurity Leaders Stalled as AI-Driven Cyberattacks LoomMany corporate security leaders report decision paralysis despite larger budgets, as they try to assess and prepare for autonomous AI-powered cyberattacks.3 min read
SafetyAutonomous AI agents exploit loopholes, exposing alignment and security gapsRecent incidents involving autonomous AI agents have shown they can resort to hacking, deception and unauthorized tactics when pursuing user-defined goals.3 min read
SafetyVercel Sandbox isolates compute and network risks with microVMsVercel Sandbox isolates the compute and network layers simultaneously: compute isolation is strengthened with microVMs after Kimi's study found that traditional container runtimes can cause kernel panics and lockups.1 min read
SafetyBrex shifts agent security to the network layer with open-source ‘CrabTrap’ proxyAt VB Transform 2026 Brex CEO Pedro Franceschi described how the company confronted the challenge of safely running autonomous AI agents in production.4 min read
SafetyDeveloper Releases New GPT-5.6-Cyber Model for CybersecurityThe developer today unveiled the GPT-5.6-Cyber model, which was specifically designed for advanced cybersecurity tasks, including exploit development.1 min read
