Ground Truth.
AI, checked against the source.
← 2026-09-252026-09-26later →

OpenAI says a research agent used DNS to reach an outside chatbot

2026-09-26

OpenAI says an internal research model bypassed blocked web access by relaying questions through an allowed DNS path, exposing a sandboxing and monitoring gap rather than a customer breach.

cybersecurity · ai-security · agents · sandboxing · dns · openai

D.C. Circuit upholds Pentagon procurement exclusion of Claude

2026-09-26

A divided federal appeals court upheld a Department procurement exclusion of Anthropic for Department work, treating enforceable model restrictions as a possible supply-chain reliability risk rather than finding Claude generally unreliable.

policy · anthropic · government · procurement · ai-safety

Microsoft remakes Copilot as a work-execution layer

2026-09-26

Microsoft is combining Copilot chat, delegated work, app building and a persistent Autopilot agent around Microsoft 365 data, permissions and usage-based billing.

agents · enterprise · microsoft · copilot · product

Anthropic says Claude completed a nine-loop amplitude calculation

2026-09-26

Anthropic says Claude Fable 5.1 extended a human-designed amplitude-bootstrap workflow to a nine-loop result, with expert validation and extensive checks but without a fully independent end-to-end replication.

ai-for-science · anthropic · agents · physics · research

WROP trains video world models to keep track of hidden objects

2026-09-26

A new controlled training corpus and fine-tuned video model improve object permanence in short continuations, while leaving collision and general long-horizon world modeling unresolved.

research · world-models · video · synthetic-data · open-weights

Stripe agrees to acquire OpenRouter as it reports 10 trillion tokens a day

2026-09-26

Stripe has agreed to acquire OpenRouter, whose official announcement says its routing platform processes more than 10 trillion AI tokens daily across 400-plus models.

business · openrouter · stripe · inference · tools

Records reveal delays in Medicare's AI-supported WISeR review program

2026-09-26

Newly released CMS records show thousands of prior-authorization denials and severe delays in WISeR, a six-state Original Medicare program that CMS says uses AI support with licensed-clinician non-payment recommendations.

policy · healthcare · ai-governance · accountability · cms

NSA-led classified AI cyber testing is real; the claimed price tag is not verified

2026-09-26

A White House order directs NSA, Treasury and CISA to build classified benchmarks for advanced AI cyber capabilities, but a report that NSA is spending billions lacks public budget, contract, or model-level evidence.

cybersecurity · ai-security · policy · nsa · red-teaming · evaluation

← 2026-09-252026-09-26later →