Using custom GPTs
OpenAI published guidance on building custom GPTs for specific tasks, focusing on workflow automation and consistent output generation.
OpenAI News·10 Apr 2026
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 published guidance on building custom GPTs for specific tasks, focusing on workflow automation and consistent output generation.
OpenAI published a general overview of applications for ChatGPT, Codex, and APIs, focusing on common use cases.
OpenAI launched a 'Financial Services' resource page, offering prompt packs, GPTs, guides, and tools for secure AI deployment and scaling.
OpenAI published best practices for safe, accurate, and transparent use of AI tools, including ChatGPT.
OpenAI published 'AI fundamentals,' a beginner's guide explaining AI, its mechanisms, and how large language models power tools like ChatGPT.
OpenAI rotated macOS code signing certificates and updated apps after the Axios developer tool supply chain attack, confirming no user data compromise.
Anthropic subjected Claude to 20 hours of simulated psychotherapy, aiming to create a more 'psychologically settled' model named Mythos.
AWS details model lifecycle management for Amazon Bedrock, outlining states, extended access, and migration strategies for evolving FMs.
AWS introduced Agent Registry (preview) within AgentCore, a centralized service for enterprises to discover, share, and reuse AI agents and tools.
Google is making Device Bound Session Credentials (DBSC) publicly available for Windows users on Chrome 146, with macOS support soon.
AWS introduced AgentCore, allowing developers to embed a live AI browser agent directly into React applications with Amazon Bedrock.
A police corporal used AI to create over 3,000 non-consensual deepfake pornographic images from women's driver's license photos.
Deep Agents Deploy is a new open-source, model-agnostic agent orchestration platform from LangChain, positioned as an alternative to Claude Managed Agents.
LangChain advocates for human-in-the-loop systems to integrate tacit knowledge into AI agents for improved performance.
AWS introduced stateful client capabilities for Bedrock AgentCore Runtime, enabling agents to request user input, generate dynamic content, and stream updates.
Revolut rolled out AIR, an in-app AI assistant offering personalized spending analysis, investment support, and biometric data checks.
Hugging Face's Safetensors and Meta's Helion joined the PyTorch Foundation, aiming to enhance security and development for ML frameworks.
Apple research proposes LaCy, an architecture for Small Language Models (SLMs) to learn by querying larger LMs for factual consistency, improving accuracy.
CyberAgent deploys ChatGPT Enterprise and Codex across advertising, media, and gaming to accelerate decisions and scale AI adoption.
Meta announced Muse Spark, a new closed-weights model, with a private API preview and public demo via meta.ai. Benchmarks show it competitive with Opus, Gemini, and GPT models.
AWS details a multi-tenant configuration pattern using key-tagging to route requests, isolate tenant data, and update configurations.
Meta's new 'Muse Spark' model released by its Superintelligence Lab, with strong general benchmarks but admitted weaknesses in agentic and coding tasks.
AWS introduced fine-tuning capabilities for Amazon Nova models on Bedrock, demonstrating improved performance for domain-specific tasks.
AWS details four human-in-the-loop (HITL) constructs for AI agents in healthcare, addressing data sensitivity and regulatory compliance.
Meta AI discusses configuration safety at scale for AI systems, using canarying, progressive rollouts, and health checks.
ALTK-Evolve introduces a framework for AI agents to continuously learn and adapt from real-time interactions, improving task performance.
Mustafa Suleyman argues AI development will continue its exponential trajectory, challenging linear human intuition about progress.
OpenAI outlines enterprise AI expansion covering ChatGPT Enterprise, Codex, Frontier model access, and company-wide AI agents.
Anthropic revealed "Mythos," a purported machine capable of discovering zero-day vulnerabilities, generating both excitement and skepticism.
Yash Tekriwal (Clay) built a custom Slack digest and Kanban dashboard using OpenAI agents and Perplexity Computer, reducing daily notifications from 150 to ~30 tasks.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion