Introducing gpt-oss-safeguard
OpenAI releases gpt-oss-safeguard, open-weight reasoning models for safety classification with customisable policy enforcement.
OpenAI News·29 Oct 2025
Use this view to inspect the underlying evidence corpus. For ranked developments, decision posture and interpretation, use Signals.
Raw feed or Signals?
Raw feed is chronological evidence. Signals ranks and interprets material change.
OpenAI releases gpt-oss-safeguard, open-weight reasoning models for safety classification with customisable policy enforcement.
Westpac is transitioning its AI deployment strategy from its compliance-focused 'AI Check' portal toward active platform orchestration.
Doppel deploys GPT-5 with reinforcement fine-tuning to detect deepfake/impersonation attacks, claiming 80% analyst workload reduction.
Microsoft and OpenAI sign updated partnership agreement expanding long-term collaboration and responsible AI commitments.
OpenAI announces recapitalization and governance restructuring, framing it as mission-aligned expansion of resources for responsible AI.
Jack Clark's Import AI #433 covers AI auditors, robotics research, and AI-managed laboratory automation.
OpenAI published a GPT-5 system card addendum detailing safety benchmarks for emotional reliance, mental health, and jailbreak resistance.
OpenAI worked with 170+ mental health experts to improve ChatGPT's distress recognition, claiming 80% reduction in unsafe responses.
Hugging Face released huggingface_hub v1.0, marking five years of their open-source machine learning platform and tools.
Datumo is presented as a new competitor to Scale AI, potentially offering faster, more cost-effective AI training data services.
Google DeepMind's Gemini 2.5 Flash-Lite, a cost-efficient model with a 1 million-token context window and multimodality, is now generally available.
Discussion on OpenArt and 'brainrot content' in AI-generated media, covering the promises and perils of algorithm-driven art platforms.
Google DeepMind's AlphaEarth Foundations integrates petabytes of Earth observation data into a unified representation for global mapping.
Mistral AI launched Mistral AI Studio, a platform offering direct API access to their models, including new enterprise-focused features.
Google DeepMind's experimental Backstory AI tool aids users in tracing the context and origin of online images.
Google DeepMind's advanced Gemini model, Deep Think, achieved a gold-medal standard on the International Mathematical Olympiad (IMO) problems.
Google DeepMind's new Perch model analyzes bioacoustic data to monitor endangered species, accelerating conservation efforts.
Google DeepMind's Gemini 2.5 Deep Think achieved gold-medal level in the International Collegiate Programming Contest, demonstrating advanced abstract problem-solving.
Nvidia is reportedly developing 'Cosmos,' a world model aiming to create smarter AI. Details are scarce, suggesting a research-stage initiative.
Google DeepMind announced an update to its Frontier Safety Framework to identify and mitigate severe risks from advanced AI models.
Google DeepMind's Gemini Robotics 1.5 enables robots with advanced perception, planning, tool use, and multi-step task execution.
Google DeepMind introduces CodeMender, an AI agent for automated identification and remediation of software vulnerabilities.
Google DeepMind partners with Commonwealth Fusion Systems (CFS) to apply AI to fusion energy research, aiming for clean energy.
Google DeepMind rolls out 'Deep Think' in Gemini app for Ultra subscribers; select mathematicians gain access to Gemini 2.5 Deep Think.
Google DeepMind released Game Arena, an open-source platform for head-to-head evaluation of AI models in competitive environments with clear winning conditions.
Google DeepMind released Gemma 3 270M, a new 270-million parameter compact model designed for hyper-efficient AI applications.
Google DeepMind enhanced native image editing capabilities within the Gemini app, allowing users to transform images with new tools.
Google DeepMind announced VaultGemma, an LLM trained from scratch with differential privacy, claiming it is the most capable such model.
Google DeepMind introduces Gemini 2.5 Computer Use model, a specialized agent-driving model for UI interaction, available via API preview.
Google DeepMind released a new 27B parameter Gemma-based foundation model for single-cell analysis, aiding cancer therapy discovery.