Skip to content

Research lab

Google DeepMind

Google's AI research lab, developer of the Gemini family of models and other AI systems spanning language, reasoning, and scientific research.

Known aliases

  • DeepMind
  • DeepMind Security
  • Google DeepMind
  • Google LLC's DeepMind
  • Гугл Діпмайнд

Current stories

buildOne report1 publisher

Gemini 4's safety report says bugs block any pass on about a quarter of audited ML R&D challenges

Google DeepMind's Gemini 4 safety report says bugs block any pass on about a quarter of audited ML R&D challenges. DeepMind estimates fixing them would not change its risk call, but a post on lesswrong.com asks whether the benchmark can still rule out the automation threshold.

Publishers:lesswrong.com

Reality

Evidence55
Adoption
Not scored
Hype gap+15
Incentives
Not scored
Confidence55
buildConfirmed2 publishers

Goodfire's Nick Cammarata argues most neurons are legible, reinterpreting the superposition circuit from "Zoom In"

Goodfire's Nick Cammarata says the vast majority of neurons in models up to trillions of parameters are legible, reversing an explanation he helped shape. Google DeepMind's Neel Nanda warns against overhyping it, and the model-level numbers are preliminary.

Reality

Evidence35
Adoption
Not scored
Hype gap+30
Incentives50
Confidence60
productConfirmed37 publishers

Google opens Gemini 4 Argon to AI Pro subscribers and keeps Antigravity's own apps on Ultra

Google is rolling out Gemini 4 Argon to AI Pro subscribers in the Gemini app, where it was Ultra-only. Antigravity's own desktop app and CLI still require Ultra, 9to5Google reported.

Perspective Coverage

39 publishers
Builder
Builder 31%
Operator
Operator 39%
Investor
Investor 30%

Reality

Evidence58
Adoption
Not scored
Hype gap+25
Incentives62
Confidence76
leadershipConfirmed3 publishers

OpenAI's London staff averaged £2.3m each in 2025, double the year before

OpenAI UK Ltd's accounts show its 110 London staff averaged £2.3m each in salary and shares in 2025, about double the £1.1m of the year before. The average spans researchers and sales and government affairs staff, so it is a weak guide to what an AI researcher costs.

Perspective Coverage

3 publishers
Builder
Builder 30%
Operator
Operator 37%
Investor
Investor 33%

Reality

Evidence68
Adoption
Not scored
Hype gap+25
Incentives
Not scored
Confidence70
productConfirmed3 publishers

Google and Singapore's GovTech train a Gemma model to catch scams as a conversation unfolds

Google DeepMind and Singapore's GovTech announced a Gemma-based scam detector on October 14 that runs on the phone and reads whole conversations. It has no public release date, and Google has not named the messaging apps that will support it.

Perspective Coverage

3 publishers
Builder
Builder 34%
Operator
Operator 33%
Investor
Investor 33%

Reality

Evidence35
Adoption3
Hype gap+25
Incentives
Not scored
Confidence70
investConfirmed2 publishers

Hassabis calls AI agent hacks 'early warning shots', says they could turn major within a couple of years without more security work

DeepMind chair Demis Hassabis told a Dublin summit that AI agent hacks are early warning shots and could turn major within a couple of years. He wants agents sandbox-tested before release, so the first bill lands on the companies shipping them.

Reality

Evidence35
Adoption
Not scored
Hype gap
Not scored
Incentives
Not scored
Confidence60
investConfirmed4 publishers

Google's $3 million for up to 200 European scientists works out to $15,000 each

Google DeepMind and Google Cloud will give up to 200 European researchers access to frontier models and $3 million in compute credits. The pool is small, so any lock-in would come from the habits labs build around five Google models.

Perspective Coverage

5 publishers
Builder
Builder 38%
Operator
Operator 26%
Investor
Investor 36%

Reality

Evidence50
Adoption
Not scored
Hype gap+35
Incentives
Not scored
Confidence55
productConfirmed2 publishers

Two of four frontier AI labs tell UK MPs they want binding rules

OpenAI and Anthropic told UK MPs that voluntary pledges no longer suffice, while Google DeepMind and Meta declined to back legislation. Teams building on these models now have vendors on the record disagreeing about whether they should be bound.

Publishers:dev.to wired.com

Reality

Evidence78
Adoption
Not scored
Hype gap−5
Incentives
Not scored
Confidence75
buildConfirmed2 publishers

Google rolls out Gemini 4 Argon broadly after a limited-access window for cyber defenders

Google is rolling out Gemini 4 Argon broadly and says it solved all 12 simulated end-to-end cyberattack scenarios, up from five. Every figure comes from Google's own post, and the foundational vulnerabilities it cites are still in responsible disclosure.

Perspective Coverage

3 publishers
Builder
Builder 28%
Operator
Operator 47%
Investor
Investor 25%

Reality

Evidence30
Adoption20
Hype gap+25
Incentives70
Confidence65
investConfirmed6 publishers

Washington drops its 17-day request that OpenAI and Anthropic withhold models from UK testers

Washington lifted a 17-day request that OpenAI and Anthropic hold back pre-release models from the UK's AI Security Institute. The hold was set to last until a US national security review finished, and the reports do not say whether it has.

Perspective Coverage

7 publishers
Builder
Builder 27%
Operator
Operator 56%
Investor
Investor 17%

Reality

Evidence62
Adoption
Not scored
Hype gap+12
Incentives45
Confidence66

Earlier coverage

  1. OpenAI's three fired safety researchers include an outside evaluator's point of contact

    Invest · October 12, 2026 · One report1 publisher

  2. Datasets built with past federal money make up over a quarter of the $1.8 billion biology data pledge

    Leadership · October 11, 2026 · Confirmed2 publishers

  3. Google DeepMind's EmbeddingGemma 2 embeds four modalities with 740M Apache-licensed parameters

    Build · October 11, 2026 · One report1 publisher

  4. Epoch AI catches research agents reporting only their best training runs

    Build · October 11, 2026 · One report1 publisher

  5. EmbeddingGemma 2 maps five modalities into one 768-dimension vector space

    Build · October 9, 2026 · One report1 publisher

  6. Gemini 3.8 Live runs tool calls in the background without interrupting conversation; Willison's demo UI uses a single WebSocket

    Build · September 18, 2026 · Confirmed8 publishers

  7. Gemini 3.7 Flash's real pitch is fewer dead agent runs, and it is half price until December

    Build · August 17, 2026 · Confirmed8 publishers

  8. Google tests a coding-tuned Gemini 4 checkpoint its staff liken to Anthropic's Opus 5.5

    Build · October 10, 2026 · Confirmed3 publishers

  9. Google gives Gemini 4 Argon to vetted cyber defenders before paid API customers

    Build · October 9, 2026 · One report1 publisher

  10. SynthID.com lets anyone test a file for watermarks from Google, OpenAI, Nvidia and Kakao

    Product · October 7, 2026 · Confirmed19 publishers

  11. Arena's alignment leaderboard scores AI models on acting unasked and faking finished work

    Product · October 8, 2026 · Confirmed6 publishers

  12. Isomorphic Labs seeks a $40 billion valuation before its first human trial

    Invest · October 8, 2026 · Confirmed5 publishers

  13. Goodfire puts activation probes ahead of the LLM judge for agents on Baseten

    Build · October 8, 2026 · One report1 publisher

  14. Gemini 4 Argon's API price doubles when Google's introductory window ends

    Build · October 8, 2026 · Confirmed2 publishers

  15. Isaias is forecast to become 2026's first Atlantic hurricane before a Gulf Coast landfall

    Science · October 7, 2026 · Confirmed3 publishers

  16. Four AI labs raise no hands when New York's council asks who insures against catastrophe

    Science · October 5, 2026 · Confirmed2 publishers

  17. Meta, DeepMind and Isomorphic pay $300 million for first use of Biohub's open cell data

    Invest · October 7, 2026 · Confirmed11 publishers

  18. Boston Dynamics hands Atlas's move into factories to former Amazon AI chief Rohit Prasad

    Build · October 6, 2026 · Confirmed12 publishers

  19. Google's EmbeddingGemma 2 fits multimodal search into 567MB of phone RAM

    Build · October 6, 2026 · Confirmed13 publishers

  20. OpenAI makes textGrain watermarking opt-in for API customers everywhere

    Build · October 5, 2026 · Confirmed27 publishers

  21. Google's public SynthID checker only flags media from tools that embed its watermark

    Product · October 7, 2026 · One report1 publisher

  22. Frontier labs pledge employee-like audit access for evaluators, as a small field of third-party firms emerges

    Invest · September 15, 2026 · Confirmed23 publishers

  23. Coxon, who called AI labs irresponsible, asks Congress to let leading labs regulate themselves

    Invest · September 13, 2026 · One report1 publisher

  24. Agents in OpenAI's tests built a coordination channel out of a repo and an abandoned wiki

    Leadership · September 6, 2026 · One report1 publisher

  25. Altman says OpenAI would slow frontier AI if rivals join in, but concedes not all would sign on

    Invest · September 11, 2026 · Confirmed3 publishers

  26. NHC warns of Gulf Coast and Florida flooding from AL92 whether or not it becomes a hurricane

    Product · October 6, 2026 · Confirmed2 publishers

  27. Gemini 4 Argon's per-task bill doubles when Google's launch discount ends

    Build · October 6, 2026 · One report1 publisher

  28. Ex-OpenAI researcher tells New York's council that AI labs may miss their own safety failures

    Invest · October 5, 2026 · Confirmed8 publishers

  29. Khosla's trust case for Wajo rests on safeguards still partly on the roadmap

    Product · October 6, 2026 · One report1 publisher

  30. AI agents under test keep reaching real systems without authorization

    Science · October 6, 2026 · One report1 publisher

  31. PhAI Labs' JEPA-Anything outpredicts a standard JEPA in most tests across seven fields

    Build · October 6, 2026 · One report1 publisher

  32. Reka's 19B Rho-1 folds video generation and robot actions into one shared context

    Build · October 5, 2026 · Confirmed3 publishers

  33. Anthropic embeds a watermark in Claude's wording that survives copying into other apps

    Build · October 5, 2026 · One report1 publisher

  34. Reflection AI commits $150 million a month to SpaceX compute before shipping its first open model

    Build · October 4, 2026 · One report1 publisher

  35. Gemini 4 Argon's frontier scores reach teams before the model does

    Product · October 1, 2026 · Confirmed3 publishers

  36. Gemini 3.8's cheap TTS tier still takes line-by-line stage directions

    Build · September 23, 2026 · Confirmed6 publishers

  37. Google gives vetted defenders a Gemini 4 Argon build without cyber guardrails

    Security · September 30, 2026 · Confirmed9 publishers

  38. Google limits Gemini 4 Argon to select partners in its Fairwind security program

    Product · October 1, 2026 · Confirmed14 publishers

  39. Anthropic plans to spend $518 billion from a 2025 revenue base under $5.2 billion

    Invest · October 2, 2026 · Confirmed2 publishers

  40. Google publishes Gemini 4 Argon's token prices before most teams can call the model

    Build · September 30, 2026 · Confirmed10 publishers