Domain and publish date filters for Web Search on AgentCore
AWS added domain and publish-date filters to Bedrock AgentCore Web Search, expanding deployment to Ireland and Tokyo regions.
AWS Machine Learning Blog·19 Aug 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.
AWS added domain and publish-date filters to Bedrock AgentCore Web Search, expanding deployment to Ireland and Tokyo regions.
OpenAI temporarily paused frontier model training to address internal cybersecurity and safety evaluation standards.
Stripe acquires AI routing and spend platform OpenRouter for $7.5B to integrate payment rails with AI API usage.
Stripe has acquired AI model routing platform OpenRouter to strengthen its position in enterprise AI gateway infrastructure.
OpenAI reaffirmed Zero Data Retention for eligible API users and previewed Private Safety Processing to perform safety checks without data storage.
NVIDIA updated its open-source FLARE framework to support federated learning workflows for multimodal vision-language models.
Anthropic is expanding its pre-IPO credit facility beyond $10 billion, up from $2.5 billion last year, according to sources.
Silicon Data is developing pricing and hedging mechanisms for AI compute to help firms manage GPU expenditure risk.
Stripe has agreed to acquire OpenRouter, a multi-model LLM routing interface and API orchestration platform.
Developers bypassed Anthropic's invisible watermarks designed for EU compliance within hours of release.
Tabular foundation models are emerging to process structured datasets where standard large language models typically struggle.
Major asset managers and Goldman Sachs partner with Nvidia on a $500B financing initiative to treat compute capacity as an asset class.
Memory hardware prices have surged 500% over the past 12 months, reversing historical cost efficiency curves to 2007 levels.
Stripe has agreed to acquire AI model gateway and token routing platform OpenRouter.
New research proposes gating untrusted RAG documents behind System 2 reasoning agents to prevent knowledge-poisoning attacks.
ArXiv paper presents six design patterns for token optimization and context management to reduce costs and latency in multi-agent workflows.
Research demonstrates that LLM gender bias evaluations are highly sensitive to minor answer format variations, complicating benchmark reliability.
Study reveals RAG reader models disagree on evidence utility signs in 33% of cases, challenging model-agnostic RAG architectures.
Research shows LLMs utilize non-human latent structures on evaluation tasks, challenging the validity of human-designed benchmarks.
Academic study reveals systematic religious bias across ChatGPT, Gemini, and Grok when generating household financial advice.
Research shows LLMs generate self-inconsistent numerical preference judgments, failing fundamental utility theory assumptions in agent pipelines.
Researchers introduced SCOPE, a framework applying conformal prediction to LLM-as-a-judge evaluation to guarantee user-defined error bounds.
Research introduces a framework measuring latent behavioral biases ('alignment signatures') across multi-agent LLM pipelines.
Research reveals LLM judges in self-evolving agent systems silently fail to retire bad skills, causing performance drift.
ArXiv paper benchmarks classical and transformer models for document sensitivity classification, highlighting label leakage risks.
A research paper demonstrates that standard MSE metrics bias irregular time-series forecasting evaluations due to timestamp sampling distributions.
Research identifies measurement flaws in agent evaluation trace analysis, demonstrating how post-response mappings distort model validity.
Research demonstrates debate-based adversarial fine-tuning reduces reward hacking in RLAIF when aligned using automated LLM judges.
Research demonstrates a novel white-box attack using knowledge editing mechanisms to force unsafe output probabilities in LLMs.
Research demonstrates Tabular Foundation Models achieve strong in-context learning transfer via self-supervised pre-training on a single real table.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion