Search signals, briefings, company results, benchmarks and glossary terms.
Search signals, briefings, company results, benchmarks and glossary terms.
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.
NIST released technical findings on the root causes of the 2021 partial collapse of the Champlain Towers South building, concluding a multi-year investigation.
OpenAI launched 'Patch the Planet,' a Daybreak initiative using AI and expert review to help open-source maintainers find and fix vulnerabilities.
OpenAI announced "Daybreak" tools, including Codex Security and GPT-5.5-Cyber, to automate vulnerability detection and patching for enterprises.
Samsung Electronics has rolled out OpenAI's ChatGPT Enterprise and Codex to its global employee base for internal use.
An op-ed argues against banning open source AI, framing it as a mistake for general, non-technical audiences.
Cloudflare has introduced Temporary Accounts for its Workers platform, allowing AI agents to autonomously deploy live serverless functions.
Datasette Apps allows hosting self-contained HTML+JavaScript applications within Datasette, enabling sandboxed SQL queries against its data.
SEC and CFTC issued a joint request for public comment on clarifying and harmonizing derivatives product definitions, impacting future regulatory scope.
Hugging Face's 'MosaicLeaks' highlights potential for AI research agents to inadvertently leak sensitive information through generated content.
OpenAI launched new spend controls and usage analytics for ChatGPT Enterprise, enabling cost management and scalable AI deployment.
The AI Now Institute warns that AI-powered labor platforms, exemplified by nursing-focused services, could expand gig-economy exploitation into new sectors.
FSB roundtable in Spain discusses external audit quality and structural shifts in the global audit industry.
The Federal Reserve Board proposes a rule for certain payment stablecoin issuers to implement effective customer identification programs.
The UK NCSC advises calibrating oversight for AI-assisted code development based on its criticality, introducing a 'vibe coding spectrum'.
The UK NCSC issued an alert urging organisations to take action due to global targeting of Fortinet firewalls and VPN gateways.
Indeed Hiring Lab reports continued payroll employment gains despite weak underlying hiring, suggesting shifts in labor market dynamics.
OpenAI claims GPT-5.5 Instant improves ChatGPT's health and wellness responses via stronger reasoning, better context, and physician evaluation.
OpenAI reasoning model identified 18 new diagnoses in previously unsolved rare genetic disease cases in children, demonstrating diagnostic aid.
Hugging Face published a blog post discussing methods and benchmarks for evaluating open-source models on agentic capabilities using custom tooling.
Hugging Face blog explores advanced fine-tuning techniques beyond LoRA, evaluating their effectiveness for specialized model adaptation.
Chinese AI lab Z.ai released GLM-5.2, a 753B parameter text-only LLM with 1M context, as open weights under an MIT license.
Google AI's AMIE, a conversational AI system, achieved primary care physician level performance in complex disease management in a Nature study.
NCSC CEO states hostile nation-states are behind three-quarters of cyber attacks targeting UK critical infrastructure systems.
BIS proposes a framework for sovereign bond portfolio construction that integrates financial and environmental considerations by modeling carbon returns as random variables.
Hugging Face is enabling agentic AI development for robot hardware via Strands Agents and LeRobot, integrating LLMs with physical systems.
OpenAI and Molecule.one demonstrated a near-autonomous AI chemist, powered by GPT-5.4, improved a drug-making reaction.
Hugging Face published a blog post on agentic resource discovery, a method for autonomous agents to search for tools and data.
OpenAI launched LifeSciBench, a new benchmark for evaluating AI systems on real-world life science research tasks, expert-authored and reviewed.
UK government collaborates with Google DeepMind on an AI prototype to accelerate housing planning decisions.
Anthropic paused its planned token-based billing for the Claude Agent SDK after user feedback, which would have significantly increased costs for power users.
The Register AI critiques the new Siri and Apple Intelligence features, describing them as cumbersome and a step backward in user experience.
AI Now's Dr. Sarah Myers West testified before the U.S. Senate Banking Committee on AI risks to the economy and public.
Google DeepMind published an AI Control Roadmap for securing AI agents, integrating traditional safeguards with real-time monitoring.
Finbarr Timbers discusses advanced post-training recipes for frontier models, covering safety, alignment, and performance optimization.
BIS research finds bond yield responses to macro news are dampened by forecast disagreement and amplified by monetary policy uncertainty.
A 'SearchLeak' exploit in GitHub Copilot allowed attackers to extract 2FA codes and other sensitive user data by manipulating context windows.
OpenAI introduced Deployment Simulation, a method leveraging real conversation data to predict AI model behavior for improved safety and evaluation accuracy.
Import AI discusses 'Alignment is not on track', 'FrontierCode', and synthetic research interns.
Expert commentary on advancements in audio reasoning, hallucination mitigation, and efficient INT8 inference for diffusion transformers.
Research claims AI governance is unprepared for an 'AGI era,' framing current AI advancements as a critical, irreversible shift.
OpenAI launched a Partner Network with a $150M investment to accelerate enterprise AI adoption and deployment through global partners.
Pyodide 314.0 now supports publishing Python packages (WASM wheels) directly to PyPI, simplifying browser-side Python deployments.
Report claims AI data centers' total water use is minimal relative to global consumption, but local impact can be significant.
OpenAI launched three new Academy courses focused on practical AI skills, workflow creation, and agent application for everyday work.
Preply uses OpenAI models to generate personalized lesson summaries and language exercises, integrating AI with human tutors.
Initial impressions of Anthropic's Claude Fable 5 highlight its "relentlessly proactive" behavior and ability to autonomously deploy multiple strategies to achieve goals.
The Federal Reserve Board finalized a rule establishing data standards for specific information collections, impacting regulatory reporting.
BIS research analyzes 593M Ethereum stablecoin transactions, revealing they are often embedded in complex, atomically executed bundles beyond simple payments.
AI Now Senior Fellow Dr. Katie J. Wells testified before the U.S. House Subcommittee on Workforce Protections regarding gig work platforms.
Bank Underground research questions if popular machine learning explainability methods truly deliver on their promises of illuminating black box models.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion