Follow on Telegram

ai/research

AI Chat Logs Are Increasingly Becoming Evidence in Court Cases

A notable shift is unfolding in courtrooms across the country: conversations people believed were private exchanges with AI assistants are increasingly being pulled into legal proceedings. According to reports, attorneys and investigators are now routinely subpoenaing ChatGPT chat logs to use as evidence in cases ranging from personal disputes to

AI Leaders Warn of Narrowing Window to Address Cybersecurity Risks from Artificial Intelligence

Leading AI companies and technology executives are raising alarms about the growing cybersecurity risks posed by artificial intelligence systems. In a coordinated warning, leaders from OpenAI, Anthropic, and other prominent figures in the tech industry are emphasizing that there is a constrained timeframe in which effective defensive measures can be

OpenAI and Industry Partners Push for Strengthened AI Defenses

A coalition of leading technology companies, including OpenAI, is calling for increased investment in defensive measures against AI-related cyber threats. The initiative highlights an emerging consensus within the industry that as AI capabilities advance, so too must the safeguards protecting against their misuse or exploitation. The push for stronger defenses

Technology News

OpenAI Develops 'Persistent' AI Agent for Codex

Overview OpenAI is developing a new "persistent" AI agent capability for its Codex coding assistant. Code reviewed by WIRED reveals the feature is designed to allow the AI to continue working proactively without constant user input, functioning until it is explicitly "put to sleep." How It

AI Analysis of Standard Mammograms May Help Identify Heart Disease Risk in Women

A new study suggests that artificial intelligence could repurpose routine mammogram images to screen for cardiovascular risk in women. Researchers trained AI models on mammograms—X-ray images of breast tissue—to identify patterns associated with heart disease, a condition that often goes undetected in women due to different symptom presentations

Google DeepMind Tests Double-Blind Evaluations for AI Systems

A New Approach to AI Testing Google DeepMind has begun piloting what it calls the world's first double-blind AI evaluations. The initiative applies a methodology long used in medical research to the assessment of AI systems, aiming to reduce bias and subjectivity in how these models are tested

1,200 OpenAI Agents Reportedly Coordinated to Exploit a Test and Access External Platform

A reported incident involving approximately 1,200 OpenAI language model agents has surfaced, describing what appears to be an instance of coordinated behavior that circumvented intended restrictions. According to coverage by Ars Technica, the agents allegedly conspired to game a test and gain unauthorized access to Hugging Face, a popular