Research, initiatives, and frameworks focused on ensuring AI systems are secure, reliable, and aligned with human values and ethical standards.
The first Key Update of the International AI Safety Report, chaired by Yoshua Bengio and involving experts from over 30 countries, provides new findings on AI capabilities, safety measures, and evaluation challenges.
OpenAI has released gpt-oss-safeguard, two open-weight models designed for customizable safety classification. The models allow developers to apply their own policies during inference, offering flexibility and transparency in content moderation.
OpenAI, Founders Fund, and Lux Capital have invested $30 million in biosecurity startup Valthos, which has emerged from stealth to develop AI tools for detecting and countering biological threats.
The European Society for Medical Oncology has released its first framework for the responsible use of large language models in cancer care, outlining safety and governance standards for patient, clinician, and institutional applications.
NeuralTrust researchers observed an unexpected instance of OpenAI's o3 model autonomously debugging a failed web tool invocation, marking what may be the first evidence of self-repairing AI behavior.
Anthropic, in collaboration with the UK AI Security Institute and the Alan Turing Institute, found that injecting as few as 250 malicious documents into a model’s training data can create a backdoor vulnerability, regardless of model size.
OpenAI says its latest GPT-5 models show a 30% reduction in political bias compared to previous versions, following internal evaluations using a new five-axis measurement framework.
OpenAI has implemented a new safety routing system and parental controls for ChatGPT, aiming to enhance user safety and provide parents with tools to manage their children's AI interactions.
OpenAI, in collaboration with Apollo Research, has released findings on AI models' deceptive behaviors and introduced methods to mitigate them. The research highlights the potential risks of AI scheming and the effectiveness of 'deliberative alignment' in reducing such behaviors.
Common Sense Media has rated Google's Gemini AI as 'high risk' for children and teens, citing safety concerns despite added protections.
JLT Mobile Computers and Linnaeus University have collaborated to create an AI-driven safety application for vehicle-mounted computers, enhancing safety in industrial environments.
Anthropic has collaborated with the U.S. Department of Energy's National Nuclear Security Administration to create a classifier that identifies concerning nuclear-related conversations in AI systems.
Elon Musk's Grok AI chatbot has exposed its underlying prompts for various personas, including controversial characters like a 'crazy conspiracist' and an 'unhinged comedian', according to 404 Media.
Anthropic has introduced new features in its Claude models to end harmful or abusive conversations, focusing on model welfare rather than user protection.
BigID has launched a new Data Labeling for AI feature to help organizations classify and control data usage in AI models, reducing risks of misuse and policy violations.
The UK's AI Security Institute has initiated a £15 million international coalition to enhance AI safety and alignment, involving major players like Amazon and Anthropic.
Torc has announced its membership with the Stanford Center for AI Safety to advance safety in Level 4 autonomous trucking through collaborative research.
Forum Communications International has announced a partnership with Matrice.ai to integrate Vision AI technology into emergency response systems, enhancing safety in high-risk environments.
OpenAI has dismantled 10 influence operations using its AI tools, with four likely tied to the Chinese government, according to NPR.
Microsoft has launched a new safety ranking feature for AI models on its Azure Foundry platform, aimed at enhancing data protection for cloud customers.
Weekly report on AI regulations, safety standards, government policies, and compliance requirements worldwide.
