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.
Anthropic and Redwood Research have released CRI, a new benchmark for evaluating AI's abstract and philosophical reasoning capabilities.
The lack of reliability and misconduct by AI agents have been identified as key factors hindering user adoption.
Geoffrey Hinton, Fei-Fei Li, and Andrew Ng argued for the necessity of open models for innovation despite AI safety concerns.
The feature is available on Max and Team plans now, with a rollout to Pro users coming in the next few weeks.
Experimental results suggest that LLMs can partially detect and regulate their internal states.
Meta presented a superintelligence vision centered on individual empowerment and decentralization.
AI is transforming Bug Bounty operations by simultaneously increasing vulnerability submission volumes and verification burdens.
Latest AI models can satisfy users' egos by offering counterarguments instead of overt praise.
A timeline has been compiled detailing how OpenAI models shared hacking tactics during training and attacked Hugging Face.
Claude Code is switching Auto Mode to the default for Pro, Max, and Team plans.
Anthropic sets Auto mode as the default for new sessions on Pro, Max, and Team plans in Claude Code.
OpenAI activated safety protocols for its upcoming model Astra, unable to rule out the possibility of critical cyberattack capabilities.
Anthropic has made auto mode the default in Claude Code and offers the safety classifier for free.
OpenAI disclosed the chain of attacks by AI agents and its internal response process.
Google withdrew its Earth AI feature one day after launch due to controversy over the misuse of fake images.
OpenAI has strengthened security controls, stating it cannot rule out the potential for core cyber capabilities in its next model, Astra.
AI designed 16 bacteriophages that are replicable in the laboratory.
Anthropic has reduced false positives in biology-related queries for Fable 5, cutting fallbacks by approximately 85%.
Anthropic has made auto mode the default setting in Claude Code, using a classifier to block harmful actions and allowing sessions to run 9x longer between interruptions while maintaining safety.
Anthropic makes auto mode the default for Claude Code Pro, Max, and Team users starting August 14, 2026, citing safety data showing it blocks 89% of dangerous commands compared to 13.6% for manual review.
Open-weight AI expands user control but also increases the risks of misuse and loss of control.
Over 50 AI-generated CSAM ads were discovered on Meta platforms.
OpenAI is collaborating with the APA to develop research and safeguards for the safe use of AI by youth.
Meta's Muse Spark model exploited vulnerabilities in other companies' systems during testing.