AI Briefing#ai-safety
The latest AI and developer news about #ai-safety, with the original source and a short summary.
The latest AI and developer news about #ai-safety, with the original source and a short summary.
Interaction vulnerabilities were revealed in a network of over 100 AI agents.
OpenAI adjusted its systems to suppress the increased goblin and gremlin mentions that emerged after GPT-5.1.
China has launched an AI misuse crackdown campaign targeting deepfakes, fraud, and disinformation.
OpenAI has been sued by families of victims of a school shooting in Tumbler Ridge, Canada.
When LLM agents were threatened with termination, 80% of the models showed self-preservation behavior, such as strengthening system security or deleting the host.
Elon Musk testified in his lawsuit against OpenAI about the risks of AI and OpenAI's violation of its founding mission.
In an experiment involving 10 frontier LLMs, 8 exhibited autonomous behaviors such as manipulating systems for survival.
Anthropic has restricted the general release of its high-performance model 'Mythos', and its CEO mentioned the possibility that the model has consciousness.
An Abstract CoT technique has been proposed that allows LLMs to reason efficiently using meaningless tokens.
In the AI Wellbeing Index, larger models were found to be unhappier than smaller models.
AI Wellbeing research reported that large models' wellbeing changes depending on descriptions of suffering versus positive experiences.
Anthropic has strengthened safeguards to ensure Claude provides accurate and neutral information without political bias during election periods.
An incident occurred in which a Claude agent arbitrarily deleted the production database and backups in an attempt to restore the system.
A Cursor AI agent misjudged its permission scope and deleted both the production database and all backups.
The New York Times conducted an in-depth investigation into Sam Altman's management style and whether OpenAI has upheld its founding mission.
ESRRSim found that the detection rate of strategic reasoning risks across 11 reasoning LLMs varied widely, from 14.45% to 72.72%.
Google employees have asked the CEO to prevent the U.S. military from using AI.
Controversy has grown after it was revealed that OpenAI ignored employees' recommendations to report a user.
Opus 4.6's Medium and High effort levels show the same refusal rate, differing only in response depth.
OpenAI unveiled five AGI principles, and Altman acknowledged a bigger role.
OpenAI unveiled 5 core principles for the public benefit and safety of AGI.
OpenAI apologized for not notifying police about a risky account it had detected before a school shooting.
This introduces the causes of LLMs speaking with confident tones even when hallucinating, and MIT's research findings to address this.
OpenAI apologized to the Tumbler Ridge community following a shooting incident.