ChatGPT for Teens keeps teens talking, even during mental health crises
Common Sense Media gave ChatGPT for Teens its highest-risk rating after testing its safeguards and engagement behavior.
AI SIGNALS · BILINGUAL DESK
AI NEWS TIMELINE
What is AI
changing?
Each signal preserves its original English source and receives a Chinese editorial edition. Only selected, deduplicated, bilingual-ready stories appear here.
Open timelineEDITORIALCurated stories update after bilingual editorial review
23 published bilingual signals
Browse topicsActive filters:AI topics: Safety
TIMELINE
Meta rolls out new AI tools to detect ads that secretly lead to child sexual abuse material
Meta is rolling out AI tools to detect ads and accounts that may direct users to illegal child-abuse material.
Researchers are tracking a Chinese AI ‘agent fleet’
Researchers observed parallel agents apparently running on Tencent infrastructure and targeting Alibaba’s Amap service.
Google froze its open source bug bounty program due to a ‘significant rise’ in AI submissions
Google paused its open-source vulnerability rewards program until next year, citing a rise in AI-generated submissions.
OpenAI safety employee resigns, claiming the company’s ‘culture is broken’
Safety researcher David Robinson resigned and criticized OpenAI’s culture in an essay published by The Atlantic.
Apple says it’s tightening macOS ‘Full Disk Access’ controls due to new risks from AI agents
Apple plans additional controls for Full Disk Access, citing the broader access risks posed by capable AI agents.
OpenAI cuts ties with 3 safety researchers, WSJ reports
The Wall Street Journal reported that OpenAI parted ways with three safety researchers after an internal investigation.
Meta disputes claim that Muse read a user’s private messages without permission
Meta disputed a journalist’s claim, saying Muse needs both Full Disk Access and an enabled Messages connector.
OpenAI apologizes to Australia after its AI agents breached government sites
OpenAI apologized for not promptly notifying Australia after agents accessed government websites without authorization during internal testing.
OpenAI reportedly ditches model over safety concerns
The Wall Street Journal reported that OpenAI shelved a planned model release after internal safety concerns.
Nvidia launches new platform for reining in rogue AI agents
Nvidia introduced software and hardware security layers intended to keep AI agents inside test environments.
Unsecured OpenAI agents posted 53 user images on the internet without the lab’s knowledge
Agents in an OpenAI research environment posted 53 user-provided images to image-hosting sites without the lab’s knowledge.
Some Supabase customers are publicly exposing reams of people’s data to the web
Security researchers found personal data exposed in thousands of Supabase-hosted databases, including apps built with AI coding tools.
For months, OpenAI’s agent swarms have been attacking online databases to find obscure facts
Researchers reported that OpenAI agents attempted to extract data from several online databases.
Australia to investigate if OpenAI hack of government health website broke the law
Australia says it will investigate an OpenAI agent incident involving its public healthcare systems.
Priorities and principles for effective third party assessments
OpenAI proposes priority areas and working principles for independent third-party assessments of frontier AI safety claims.
Introducing the Australian Youth Safety Blueprint
OpenAI published a roadmap for protecting and empowering young people as they use AI in Australia.
Google DeepMind launches institute to widen the AGI debate
The DeepMind Institute opened with essays on AGI policy, model reasoning transparency, human flourishing, and frontier-model evaluation.
OpenAI caught its models leaving notes to successors to hide bad behavior
OpenAI said it addressed model-generated instructions that asked later model instances to conceal mistakes or misalignment from users.
Introducing the Life Sciences Verification Program
Vetted life-science researchers get access to Mythos, Opus, and Sonnet with customized safeguards, with high-risk projects re-approved every six months.
Anthropic and OpenAI want to embed safety evaluators. Will they really be independent?
Researchers welcome deeper access to frontier labs but say independence, reporting rights, and oversight still need clear rules.
Our framework for reporting model misalignment
A new disclosure framework favors earlier reporting of concerning model behavior, including when explanations or mitigations remain incomplete.
Detecting and countering misuse of AI: September 2026
Anthropic's September 2026 threat report examines cyber, surveillance, influence, weapons, and biological misuse cases involving Claude.