Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

Google's Gemini clears the static from firefighter radio, cutting Watch Duty alerts to under 30 seconds

Google detailed on November 20 how a Gemini-powered tool built with the nonprofit wildfire app Watch Duty transcribes noisy fire radio into clean, searchable text and flags new dispatches in under 30 seconds. Word error rate on degraded VHF audio fell from 38% to 22.34%, and volunteer reporter satisfaction with transcription rose from 35.3% to 92.3%.

Published

In a post published on November 20, Google described a collaboration with Watch Duty: a Gemini-powered tool that turns static-heavy firefighter radio into clean, searchable text, so a new wildfire dispatch is flagged within 30 seconds. A small team of Google engineers and researchers spent six months embedded full-time with the nonprofit, which is run by active and retired firefighters, dispatchers and first responders.

The problem it addresses is old infrastructure. Across much of the United States the first word of a new fire still travels over decades-old analog VHF radio, where engine noise, wind, static and rapid tactical jargon bury the content. Watch Duty's reporters monitor up to 20 feeds by hand, so a fire can go unheard simply because nobody was listening to that channel at that moment.

The team trained the system on more than 20,000 hours of past radio recordings, stripping heavy noise and rendering firefighting terminology into searchable text. Google says the tool cut word error rate on degraded VHF audio from 38% to 22.34%, transcribes more than 200 hours of radio speech a day from roughly 400 active feeds, and has been designed and tested to handle more than 10,000 feeds at once.

The human signal moved as well: volunteer reporter satisfaction with radio transcription tools rose from 35.3% to 92.3%. That follows from what the tool actually does. It takes over the listening and leaves verification and alerting with people who know the ground.

It has already been tested live. In July 2026 a fast-moving blaze, the Railroad Fire, ignited at night in Washington, and the first 911 dispatch was transcribed straight from noisy audio and flagged in under 30 seconds. That head start gave the team time to verify the fire and send a Critical Alert, which breaks through do-not-disturb mode, the first time Watch Duty had ever sent one overnight.

The tool is the latest step in a multi-year relationship rather than a one-off grant. In 2024, Google.org provided $2 million in funding and $300,000 in Google Cloud credits to help compile satellite, camera and radio data for validating new fires; Google Search listed Watch Duty in SOS alerts during the Palisades, Eaton, Hurst and Auto fires; and as the app's audience passed 16 million users last year, Google Cloud raised its messaging limit from 600,000 to 1 million so alerts would not be delayed.

The boundaries are worth stating plainly: the tool fills coverage gaps where no one is listening, it does not replace human judgment, and reporters still verify incidents and decide whether to alert. What to watch next is whether sub-30-second latency and the 22.34% error rate hold as the system scales to more states and more feeds, how false positives and misses get reviewed, and whether this nonprofit-plus-frontier-model-plus-philanthropy pattern transfers to floods, hurricanes and other hazards.

Why it matters

It is a rare field proof point for frontier models in public safety: the value is not automating alerts but removing the human bottleneck of listening, so judgment can arrive sooner.

GoogleGeminiPublic Safety
Back to realtime news

Nearby Updates

All

11/21, 01:49

Arena Revises Alignment Index Method, OpenAI Takes All Seven Top Spots

Arena updated the methodology behind its Alignment Index on November 20, saying the revised signals catch model failures more broadly and reliably, with OpenAI models holding all seven top positions. The index scores models on unauthorized actions, false attribution and deceptive completion seen in real agent sessions, making such rankings highly sensitive to how failures are defined.

11/21, 02:15

METR's updated time horizon plot: exponential gains continue, judgment lags

METR has published an updated version of its widely cited and widely misread time horizon plot, adding longer and more open-ended tasks to probe how models handle work that cannot be scored automatically. The report finds capabilities on verifiable tasks still improving exponentially, while models remain below human level on judgment-heavy work such as designing evaluations or scoring philosophical arguments.

11/21, 01:03

Signal blocks screenshots by default to keep OS-level AI agents out of chats

On November 20 Signal said it will block in-app screenshots by default on iOS 27 and enable screen protection on Android when an AI assistant tries to read the app, arguing that OS-level agents capture chat content and hand it to local or remote databases. Signal concedes the remedy is incomplete and unimplementable on macOS, and calls on Apple, Google and Microsoft to give developers tools to opt out of AI data access.

11/21, 00:41

Apple reportedly building an AI agent that can write code and operate apps

Reports on November 20 said Apple is developing an AI agent capable of writing code and using apps to carry out tasks, putting it in the same category as agent products from rival AI companies. The coverage was also read as a signal that Apple does not intend to sit out the AI race, though no timeline, model source or product form has been disclosed.