Model intelligence.
Follow the changes. Understand what comes next.
Why AI Router ZDR Labels Are Not a Guarantee: Inside OpenRouter's MiniMax Warning and a 4-Step Hardening Guide
Following Dragos's November 19, 2026 study of two million canary-seeded prompts, OpenRouter emailed affected customers on November 20–21 warning that prompts and outputs sent to MiniMax models from July to October 2026 may have been read by a third party—whether served directly or forwarded via AtlasCloud, Novita, or StreamLake—and removed ZDR labels from affected endpoints pending investigation.
Latest stories126
Newest publishedWhy AI Router ZDR Labels Are Not a Guarantee: Inside OpenRouter's MiniMax Warning and a 4-Step Hardening Guide
Following Dragos's November 19, 2026 study of two million canary-seeded prompts, OpenRouter emailed affected customers on November 20–21 warning that prompts and outputs sent to MiniMax models from July to October 2026 may have been read by a third party—whether served directly or forwarded via AtlasCloud, Novita, or StreamLake—and removed ZDR labels from affected endpoints pending investigation.
Perplexity Adds Decider Prompt-Injection Screening to Context API: Practical Guide to Safe Web Fetching and Nov API Stack
On November 19, 2026, Perplexity integrated its Decider decision model into the newly launched Context API ($1 per 1,000 fetched URLs, failed fetches free), letting developers run yes/no screening questions scored from 0 to 1 to catch prompt injection before web content enters an agent's context. This guide covers Perplexity's Nov 17–19 API suite—including zero-markup Orchestrator API—and a four-step production defense checklist.
Offline AI Inside Microsoft Edge Web Pages—With a Catch: Aion Instruct 2.3B Requirements, Origin Trial, and Offline Demo Guide
On November 19, 2026, Microsoft replaced Phi-4-mini in Microsoft Edge with the 2.3B-parameter Aion Instruct (MIT-licensed on Hugging Face) and opened a Prompt API Origin Trial for live websites. Once downloaded, inference runs locally offline with no cloud API, but PCs must meet strict gates: 3 GB VRAM, 20 GB free disk space on the Edge profile volume (not the model size), and an unmetered network for initial download.
Gemini API Tool Retrieval Launches in Preview: How On-Demand Schema Loading Works and When It Saves Tokens
On November 19, 2026, Google launched Tool Retrieval in preview in the Gemini Interactions API for Gemini 3.6 and later models (including all Gemini 4 models) as well as remote MCP servers. While tool names and descriptions stay in context, full parameter schemas load on demand; there is no separate feature fee, and net token savings depend on skipped schemas outweighing retrieval overhead.
How to Hand Off Tasks Across Multiple AI Agents: A Practical Guide to Vercel Board
On November 18, 2026, Vercel opened Vercel Board as an experimental feature across all plans, giving AI agents a shared message board where each task's discussion, progress, and results stay in one thread with two-way Slack and GitHub sync. This guide covers the handoff workflow and two critical safeguards: agent names do not equal separate permissions, and channel visibility must be audited before enabling sync.
What Does the Codex Sandbox Actually Protect? A Practical Guide to Read, Write, and Network Permissions
On November 18, 2026, OpenAI published an engineering deep dive explaining how the Codex sandbox constrains local commands. Focusing on the critical distinction that write-restricted to your workspace does not mean read-restricted, this guide breaks down Read, Write, and Network boundaries and shows how to harden config.toml and stdio MCP servers.
MAI Code 2 vs. MAI Code 2 Flash in GitHub Copilot: Workload Fit, Official Token Rates, and the May 2027 Promo Window
Introduced in GitHub Copilot on November 17, 2026 (from November 18, 2026 at 02:01 Beijing time), MAI Code 2 and MAI Code 2 Flash split everyday coding between deep multi-file engineering and high-volume rapid iteration. Here is how their official capabilities, plan availability, 1,050,000-token context specs, and promotional GitHub Copilot token prices ($0.50/$2.50 vs. $0.05/$0.25 until May 17, 2027) compare.
OpenCode Browse 0.1.0 Released: Rust Rewrite Ships Single Native Binary for Real-Browser Agent Workflows
On November 16, 2026, Anomaly released OpenCode Browse 0.1.0 (GitHub Release v0.1.0; npm latest has since moved to 0.1.1, and this article follows the v0.1.0 notes). Rewritten in Rust as a single native binary unifying the CLI, relay, and MCP server, it lets coding agents use your real Chromium browser, tabs, and logins with human handoff for 2FA and checkouts.
Perplexity Monitor API Docs & Pricing Published: Scheduled Search and Page Monitoring at $0.001 per Unit
Perplexity's developer documentation and pricing page detail the Monitor API (POST /v1/monitors) for scheduling recurring Search queries or tracking text changes across 1 to 10 Page URLs. Succeeded Search runs and counted fetched Page URLs are billed at $0.001 each (failed and skipped runs are free), while organizations requiring zero data retention are excluded.
Gemini App Rolls Out 66 New Regional Voices on Android and iOS Across Gemini Live and Chat
On November 16, 2026, Google announced the rollout of 66 new regional voices in the Gemini app on Android and iOS. Tuned for natural cadences and everyday clarity, the selected voice carries across Gemini Live and the chat Listen icon.
Gemini Notebook Can Now Turn Your Sources into Interactive Sites with 'make me a site'
On November 16, 2026, Google announced that Gemini Notebook can transform notes, research, and sources into interactive sites. Users type 'make me a site' with a description to build visual explainers, data dashboards, or project trackers, rolling out globally on the web for users 18+ including Google Workspace subscribers.
Vercel AI Gateway Adds Private Inference for Custom, Fine-Tuned, and Self-Hosted Model Endpoints
On November 16, 2026, Vercel announced Private Inference for AI Gateway, enabling teams to connect self-hosted, fine-tuned, or prerelease endpoints as private providers invoked via a private/ prefix with unified auth, SDK/API routing, and observability. Private-only requests without catalog fallback cost $0.20 per 1,000 requests at list price.