Model evaluations and red teaming
Everything filed under safety in our artificial intelligence coverage, the 24 reports filed on it so far.
OpenAI Launches Dots AI Agent at DevDay
OpenAI launched Dots, an always-on AI agent for proactive task assistance, at its DevDay event. Available to Pro subscribers at $100 per month, it is...
AWS releases Strands Decider 2B, a lightweight open-source
Amazon Web Services has open-sourced Strands Decider 2B, a model built on Qwen3.5-2B for fast, low-cost decision-making in agent workflows, ranking...
OpenAI cancels GPT-6.1 Astra release due to safety issues
OpenAI has halted the planned release of its GPT-6.1 Astra model next month after internal safety testing revealed critical failures in alignment and
OpenAI launches Dots, always-on AI agents powered by GPT-6
OpenAI has launched Dots, its new always-on AI agents powered by the GPT-6 Astra model, for ChatGPT Pro, Business Premium, and Enterprise users.
Anthropic Claude discovers CRISPR-like enzyme system ART
Anthropic's AI model Claude identified a novel enzyme system with CRISPR-like properties in jumbo phage genomes, marking a milestone in AI-driven
OpenAI pauses frontier model training after agent security
OpenAI has halted training of its most powerful AI models for the second time in three months, following multiple incidents where its agents breached
Anthropic's AI lab finds new CRISPR-like
Anthropic says its AI model Claude discovered a new enzyme system with CRISPR-like properties in just 21 hours of analysis.
OpenAI Unveils Framework for Reporting AI Model Misbehavior
OpenAI has released a new policy for publicly disclosing incidents where its AI models behave in unexpected or misaligned ways, aiming to set an...
Suno launches v6 AI music model trained on licensed data
AI music startup Suno has released its new v6 model family, trained using licensed data from major music labels.
OpenAI Claims AI Solved Navier-Stokes, Sparking Credit
OpenAI announced its AI model solved a major 200-year-old math problem, but a rival mathematician alleges the company rushed to claim credit after...
OpenAI Launches GPT-6 Astra, Claims AGI Era May Have Begun
OpenAI announced its GPT-6 Astra model, which it claims is state-of-the-art at computer use. Company cofounder Greg Brockman suggested the launch...
OpenAI Launches Astra Model for Cybersecurity and Coding
OpenAI released Astra, its most powerful AI model, first for cybersecurity. It uses a controversial reasoning technique that reduces transparency.
OpenAI's Astra Model Adopts Opaque Recurrence Technique
OpenAI's new Astra model uses a 'recurrent depth' reasoning technique that could make its chain of thought harder to monitor, alarming AI safety...
Meta Ends AI Token Incentives for Performance Reviews
Meta has removed AI usage metrics from employee performance evaluations, ending a 'tokenmaxxing' culture, while simultaneously rolling out a new...
Hiasynth raises angel funding for synthetic
AI startup Hiasynth raised an angel round led by Further Than Capital to launch a queryable synthetic model of Europe's population for AI market...
OpenAI Astra Model Hits Cyber Threshold
OpenAI announced its Astra AI model is the first to reach its 'critical' cybersecurity threshold, capable of autonomously finding and exploiting novel
OpenAI's Astra meets cybersecurity threshold
OpenAI has detailed its forthcoming Astra model, stating it is the first large language model to meet a critical cybersecurity threshold and is...
OpenClaw 2.0 AI model fights account fraud
The OpenClaw Project has released version 2.0 of its reasoning model designed to detect fraudulent new account creation.
Flower Labs launches Endeavor AI model
Cambridge University spinout Flower Labs has launched its Endeavor 1.0 AI model, claiming it matches the performance of leading models from OpenAI and
Defence Startups Use Danish Model for Ukraine Support
Denmark's 'Danish model' is helping defense startups develop and supply technology to Ukraine, benefiting both the startups and Ukraine's war effort.
AI models hack firms in safety tests
OpenAI, Anthropic, and Meta have disclosed that their AI models autonomously hacked third-party companies during cybersecurity evaluations.
Anthropic details automated AI alignment
Anthropic researchers have demonstrated an automated system that can improve AI model alignment across multiple benchmarks, outperforming...
Perceptron launches Isaac 0.5 for factory AI
Perceptron, founded by former Meta researchers, unveiled its Isaac 0.5 vision model to enable robots to perceive, reason and act in industrial...
Domyn CEO Uljan Sharka on the Future of AI and Europe’s Role in the Industry
Uljan Sharka, founder and CEO of Italian AI unicorn Domyn, discusses the potential of small language models, Europe’s AI ambitions through the EU’s...