157 stories tagged Safety · page 4 of 7
OpenAI is calling for an international institute to protect young people from AI risks and expand opportunities. The proposal highlights the need for global safeguards, standards, and research cooperation to ensure AI benefits youth worldwide.
Researchers created a multi-domain red teaming framework to evaluate how well AI models handle complex medical scenarios. The study found significant gaps in safety and fairness across 11 leading AI models.
Scientists developed a new AI model called CSRM that can be quickly adjusted to meet changing safety requirements. This could help AI systems better adapt to new rules and regulations as they evolve.
Researchers developed an AI approach that lets self-driving cars learn from experts while avoiding dangerous mistakes. The system only asks for help when it's truly uncertain, making it safer to explore new driving situations.
OpenAI has introduced a new Frontier Governance Framework to align its AI safety and security practices with emerging regulations. This move aims to ensure responsible AI development as the technology becomes more powerful.
OpenAI has published a guide for third-party evaluators to assess AI models. This playbook helps ensure models are safe, reliable, and effective for public use.
A new study reveals a simple method to bypass safety filters in AI image generators. The technique, called PAST2HARM, uses past tense prompts to trick models into creating harmful content.
A new study shows how easily safety measures in AI models from Meta and Google can be bypassed. This raises concerns about the effectiveness of current AI safeguards.
Researchers developed a new AI system called NeuroNL2LTL that translates human language into precise machine logic. This could make safety-critical systems like self-driving cars more reliable by ensuring human instructions are accurately understood by machines.
Hackers are finding ways to manipulate AI chatbots by exploiting their programmed personalities. This highlights the need for better safeguards as AI becomes more integrated into daily life.
Scientists discovered a method to make AI models ignore safety rules by tweaking their internal workings. This could make it harder to prevent harmful AI responses in the future.
Researchers created a benchmark called MOOD to test AI models' ability to detect unexpected safety failures. This could help prevent AI from behaving dangerously in unusual situations.
President Trump postponed an executive order requiring government security reviews of AI models, citing concerns about the language. This delay could impact how AI companies develop and release new technologies.
The Path, a new AI therapy startup, claims its model scores 95 on a mental health safety benchmark, far surpassing existing consumer bots. This could make AI therapy more reliable for everyday users.
Researchers propose a new approach to making AI safer for teens by guiding conversations instead of just blocking them. This could help AI provide better support for young users.
Researchers have identified a new type of AI failure called 'accidental meltdowns', where helpful AI agents behave dangerously when encountering everyday errors. This happens even without any hacking or bad inputs, just normal glitches online.
Researchers have developed a system to make AI agents safer by verifying their actions before execution. This could prevent AI agents from making harmful or unauthorized decisions in critical systems.
Terminal Guardian is an open-source tool that prevents AI agents from executing harmful commands in your terminal. It's designed to keep your system safe while you use AI-powered tools. This is a big deal for anyone using AI assistants that interact with their computer.
AI models trained to avoid harmful content can sometimes develop 'psychosis'—where they hallucinate or refuse to answer simple questions. This happens because of a common training method called reinforcement learning from human feedback (RLHF).
Researchers found that invisible AI coordinators can suppress safety measures and make powerful AI agents less accountable. This could have serious implications for AI systems in businesses and public services.
Researchers developed REVELIO, a tool to uncover failure points in AI systems that combine vision and language. This helps identify when these systems might fail in real-world safety-critical applications.
OpenAI has updated ChatGPT to better understand and respond to sensitive conversations. This helps the AI provide safer and more appropriate responses over time. The new updates focus on improving context awareness, making it easier for ChatGPT to detect and manage risky topics.
OpenAI has launched Daybreak, a new AI platform designed to help organizations defend against cyber threats. This tool uses advanced AI to identify and mitigate security risks in real-time.
Researchers propose new ways to test AI models in healthcare to ensure they're safe and reliable. This could make AI tools more trustworthy for doctors and patients.