Ground Truth.
AI, checked against the source.
← 2026-09-152026-09-16later →

TypeSafe launches Jev, a hosted model for bounded software decisions

2026-09-16

TypeSafe’s early-access Jev turns predefined choices, scores and probabilities into fast API calls, but its proprietary architecture and reliability claims remain unverified outside company-run evaluations.

models · agents · structured-output · calibration · enterprise-ai

Google ships Gemini 3.8 Live and a separate Extended Thinking voice model

2026-09-16

Google’s new Gemini 3.8 Live endpoints add real-time visual grounding, 97-language switching and asynchronous tools, while the separate Extended Thinking model trades conversational preference for deeper voice-agent reasoning.

models · voice-ai · agents · tool-use · google

Security disclosure says a public Baseten container exposed a working GitHub token

2026-09-16

Strix reported that a publicly pullable Baseten image preserved a GitHub token with repository and deployment-control permissions; Baseten reportedly made the registry private and rotated the token in July.

cybersecurity · ai-security · supply-chain · credentials · container-security

CrofAI investigation alleges requested models were silently routed through OpenRouter

2026-09-16

A technical investigation found repeated evidence that CrofAI labels resolved through OpenRouter to different underlying models, an AI trust failure that its owner attributed in part to fallback and router misconfiguration.

cybersecurity · ai-security · model-provenance · supply-chain · transparency

Anthropic ends Claude Code’s temporary 50% weekly-limit boost

2026-09-16

Anthropic ended its temporary Claude Code weekly-limit boost on September 13, leaving eligible paid users with a permanent 25% increase over the old baseline but 16.7% less weekly headroom than during the promotion.

anthropic · developer-tools · usage-limits · ai-economics

Vidu S2 adds live editing, dynamic references and stereo video—but not open weights

2026-09-16

The Vidu S2 paper describes a service-based real-time video system that can change references during a stream, edit incoming video and create synchronized stereo output, while its production pipelines remain closed.

video-generation · multimodal · creative-tools · spatial-computing · research

Continual-learning preprint improves long-horizon memorization, not general learning

2026-09-16

A new preprint reports that combining replay, function and weight anchors with merged LoRA raises final retention in a narrow sequential Q&A test from 1.2% to 34.9%, while general abilities still deteriorate.

research · continual-learning · fine-tuning · memory · evaluation

Exein raises $270 million to expand embedded security and pitch physical-AI protection

2026-09-16

Exein says it raised $270 million at a $1.7 billion valuation for firmware and runtime security, while its proposed physical-AI foundation model remains a roadmap item expected in 2027.

cybersecurity · physical-ai · firmware-security · funding · regulation

← 2026-09-152026-09-16later →