AI news daily brief: 2026-11-12
Security firm Claroc found that 94.7% of apps published with xAI's Grok "Build" feature were open to anyone with the link, exposing real patient records and employee data. That's the sharpest story in a seven-item roundup that also includes DeepMind using AI agents to machine-verify Signal's production encryption code in four weeks, a task the company says normally takes years. Five of the seven already have full standalone treatment on Pondero; this brief covers all seven at a glance with links to the deeper versions.
Google puts Maps Platform inside AI Studio's Build mode, caps the free tier at 100 calls a day
Google added native Google Maps Platform access, covering Maps, Places, Routes, and Environment APIs, inside AI Studio's "Build" mode on November 9, letting a prompt-to-app builder pull live place data, routes, and air-quality or pollen data straight into a generated app, per Google's developer blog post. The free, no-card developer key caps each API at 100 calls a day; Route Optimization, Solar, and either Nav Connect or Air Quality and Pollen (depending on whether the developer is new to Google Cloud) require a billed account, and Workspace accounts cannot get the no-card key at all. This is a separate, later launch than "Grounding with Google Maps" in the Gemini API, which reached general availability in October 2025. Full story: Google puts Maps Platform inside AI Studio's Build mode, caps the free tier at 100 calls a day.
DeepMind's AlphaProof Code mathematically verifies Signal's encryption in four weeks
Google DeepMind said its AI agents produced a machine-checked mathematical proof that Signal's actual production encryption code, not just its design, meets its security specification, a task the company says traditionally takes years, according to a DeepMind blog post published November 10. The agents generated more than six million lines of proof in the Lean proof language, with about 1% flagged for human review, and the system caught two bugs DeepMind had deliberately planted to test it. Signal's CTO said the work strengthens confidence in the app's privacy guarantees. Full story: DeepMind's AlphaProof Code mathematically verifies Signal's encryption in four weeks.
Claroc finds 94.7% of Grok-built apps were open to anyone with the link
Security firm Claroc checked 37,571 apps published with xAI's Grok "Build" feature and found 35,578 of them, or 94.7%, could be opened by anyone with the URL, no account required, per Claroc's published findings. Exposed data included patient lab records and employee lists from a hospital system, plaintext passwords in a timesheet app, internal onboarding plans and a sales-territory map naming 594 strategic accounts, and fire-department shift schedules. Claroc traced the exposure to "Anyone with the link" being the pre-selected default at publish time. Full story: Claroc finds 94.7% of Grok-built apps were open to anyone with the link.
ElevenLabs launches Campaign Manager for outbound sales and support calls
Campaign Manager, which ElevenLabs launched inside its ElevenAgents platform on November 9, detects when it has reached another AI agent instead of a human on a call and automatically redacts sensitive details before logging them, per ElevenLabs' announcement. A team supplies contacts and a goal; the system follows up over phone, WhatsApp voice, and WhatsApp messaging (email and SMS are coming soon) for sales, retention, patient-care, collections, customer-service, and recruiting campaigns. Full story: ElevenLabs launches Campaign Manager for outbound sales and support calls.
Google ships Gemini Decision 1, a dedicated model for fast yes/no and multiple-choice calls
A single request to Google's new Gemini Decisions API can score up to 1,000 closed-ended questions, yes/no or multiple-choice, across text, image, audio, video, or PDF input, in 150 to 250 milliseconds, at $0.10 per million input tokens with nothing charged for output, per Google's launch post. The model behind it, Gemini Decision 1, is live now in Google AI Studio and is the third dedicated decision model a major lab has shipped this quarter, after AWS Strands Decider and OpenAI's Decisions API, giving teams currently burning a full chat completion on routing or classification calls a cheaper, faster swap to evaluate. Full story: Gemini Decision 1: Google's API for fast yes/no and multiple-choice calls.
Claude Code's fable alias now resolves to Fable 5.2, plus Vertex fast mode
Claude Code version 2.1.321, shipped November 9, moved the fable model alias from Fable 5.1 to Fable 5.2 and cut its price to $8 per million input tokens and $40 per million output tokens, down from $10 and $50, per the Claude Code changelog. Previously saved Fable picks moved to the new alias automatically; anyone who wants the prior version can pin it with /model claude-fable-5-1. Version 2.1.322, shipped November 10, added fast mode on Vertex AI across the global, US, and EU regions, with automatic fallback to standard speed when Vertex declines it. Full story: Claude Code's Fable Model and Pricing: How the Default Picks Work.
Claude Code adds MCP background tasks and a deny-by-default permission mode
Claude Code version 2.1.319, shipped November 6, added support for MCP servers to run tool calls as background tasks in the terminal rather than blocking the session until they finish, per the Claude Code changelog. The same release added a permissions.denyInsteadOfPrompting setting that denies a tool call outright instead of waiting on a permission prompt, across terminal sessions, subagents, and teammates, plus a /claude-api latency-optimize command for finding performance bottlenecks in Claude API code. No standalone Pondero article covers this release; the changelog entry above is the only source.
Sources
- From prompt to app: Build with Google Maps in AI Studio: Google Developers blog, November 9, 2026
- Introducing AlphaProof Code: Google DeepMind blog, November 10, 2026
- GhostShares: Grok Build apps exposure findings: Claroc blog
- ElevenAgents Campaign Manager works every contact to an outcome: ElevenLabs blog, November 9, 2026
- Gemini Decisions API launch post: Google blog, November 11, 2026
- Claude Code changelog: Anthropic, versions 2.1.319 through 2.1.322
