Google AI Updates: April 29, 2026

1. Google Picks Up the Pentagon AI Contract Anthropic Refused

Google. TechCrunch reports that Google signed a new Department of Defense contract giving the Pentagon expanded access to its AI for classified work, days after Anthropic publicly refused to permit the same use cases — domestic mass surveillance and autonomous weapons — under its usage policy. The move, signed in the face of objections from over 600 Google employees, marks a sharp divergence between the two labs on defense exposure: where Anthropic has hardened its policy to opt out of specific government applications, Google is leaning further in. The contract value and scope weren’t disclosed, but the strategic signal is unmistakable for any agency picking a primary AI vendor for sensitive workloads. Source

2. YouTube Tests an AI Search Feature That Combines Text Answers, Full Videos, and Shorts

Google. YouTube began rolling out an AI-powered search experience to US Premium subscribers on an opt-in basis, returning a conversational guided answer alongside a curated set of full-length videos and Shorts in response to a query. The feature builds on Google’s Search AI Overviews work but applies the same answer-engine pattern to a video-first surface — closer in spirit to Perplexity’s video-grounded answers than to a pure SERP. For creators, it changes the unit of discovery from “the video that matches the search query” to “the segment the model decides to cite,” which has obvious implications for view-time accounting and source attribution that Google hasn’t yet addressed publicly. Source

3. Google Translate Hits 20 Years, Now Spanning Almost 250 Languages

Google. Google marked Google Translate’s 20th anniversary with a retrospective covering its arc from a 2006 statistical-MT experiment to a near-250-language neural system — including the recent additions powered by Google’s PaLM 2 and Gemini-derived translation stacks. The post is a marketing artifact, but the headline number matters: the addition of low-resource languages over the last two years (many of them via Google’s “1,000 languages” initiative) is one of the most concrete examples in the industry of large-model capability spillover into a previously underserved domain, and a useful benchmark when evaluating any new “all the world’s languages” translation pitch from competitors. Source