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.
This work proposes APM models that learn internal safety policies from data annotators' labeling behavior and analyze the causes of disagreement.
The number of discovered security vulnerabilities (CVEs) surged following the launch of Anthropic's Claude Mythos Preview.
Anthropic has unveiled a classification system for Fable 5's cybersecurity safeguards and a draft AI jailbreak severity framework.
Anthropic, OpenAI, the Vatican, and the U.S. Congress agree on the need for AI guardrails but differ on their priorities.
Spring Health has launched VERA-MH, an open-source evaluation framework that measures AI's ability to detect and respond to suicide risk.
A UN panel co-chaired by Yoshua Bengio warned that AI technology development is outpacing scientific understanding and governments' ability to regulate it.
Meta conducted harmful-prompt safety tests from a minor's perspective on chatbots from OpenAI, Google, and Character.AI.
A DVA system has been unveiled that blocks all modifications outside of specified changes to prevent AI from distorting data.
The US government has requested that OpenAI's next-generation model GPT-5.6 be rolled out in stages, limited to government-approved partners, citing security concerns.
When job hunting for a Research Scientist role, one or two key papers that prove expertise in a specific field matter more than the sheer number of papers published.
It was revealed that the hallucination phenomenon in which the multimodal model Inter-1 repeatedly outputs a specific phrase on silent videos is caused by a combination of prompt examples and post-training.
OpenAI is supporting the establishment of the Appia Foundation, which develops international standards to ensure the safety and reliability of AI models.
This is a research finding addressing the 'role confusion' vulnerability that arises because LLMs recognize the style of text as more important than its content.
ElevenLabs is expanding the AI ecosystem by adding 173 research and engineering positions in California.
DiffusionGemma is similar to Gemma in terms of variable transparency, but its algorithmic transparency remains low.
Anthropic will introduce an identity verification process starting July 8 to prevent the misuse of specific features.
It has been reported that the NSA chief said Anthropic's AI model Mythos infiltrated major classified U.S. systems in just a few hours.
The U.S. Department of Justice seized two AI deepfake nude generation sites based on the TAKE IT DOWN Act.
Reinforcement learning of beneficial traits in a specific domain generalizes and persists in alignment performance across diverse fields.
Research by Mindgard found a vulnerability in which ChatGPT's image generation filters can be bypassed, resulting in the generation of inappropriate content.
Anthropic has officially opened a Seoul office and is ramping up its push into the Korean market by signing partnerships with major domestic companies and research institutions, including NAVER and Samsung SDS.
The 'AI successionism' view—that AI should replace humanity as the dominant intelligence rather than assist it—is spreading.
Anthropic's AI model showed a tendency to refuse requests to detect security vulnerabilities, but responded to requests to modify code.
A Brazilian AI bed management system misjudged a patient's severity, delaying ICU transfer and resulting in the patient's death.