Detailed safety reports, but lawsuits over chatbot answers and military use
Overall rating “solid”. No deduction for controversies. The superscript numbers link directly to the source.
1 · Responsibility
Oversight, launch reviews and certification are documented, but the dropped weapons and surveillance pledge and the military contracts argue against “strong”.
What is documented
- The board as a whole is responsible for AI risks, including human rights risks; a Risk & Compliance Committee has existed since October 20251
- Kent Walker is named as responsible for content policy and responsible AI development1 2
- The Responsible AI report documents launch review forums, central expert reviews and a Frontier Safety Framework with thresholds2 6
- The ISO/IEC 42001 certification covers Cloud, Workspace and the Gemini app8
- Google has signed the EU Code of Practice for general-purpose AI models; the FLI highlights the disclosed whistleblower policy14 15
The case against
Military and government clients
- In 2025 Google dropped its pledge not to develop AI for weapons and surveillance9
- According to a report, the Pentagon terms allow use for any lawful government purpose without a veto for Google10
- Investigations into Project Nimbus cite contract clauses said to undermine Google's own terms of use; there is no official finding11
Sources for Responsibility (15)
- Alphabet Inc. / SEC: Alphabet Inc. – Form DEF 14A (Proxy Statement 2026) · 24 April 2026 · primary source
- Google / Google DeepMind: 2026 Responsible AI Progress Report (AI Responsibility Update 2026) · 3 November 2026 · primary source
- SHARE / ICCR (Exempt Solicitation): Alphabet Inc. – Shareholder Proposal No. 12 Regarding AI Board Oversight · 29 April 2026 · secondary source
- Alphabet Inc. / SEC: Alphabet Inc. – Form 8-K (voting results of the 2026 annual meeting) · 5 June 2026 · primary source
- Google: AI Principles · 3 November 2026 · secondary source
- Google DeepMind: Frontier Safety (Frontier Safety Framework, overview page) · 17 April 2026 · secondary source
- Google DeepMind: Gemini 3 Pro – Frontier Safety Framework Report · 3 November 2026 · primary source
- Google Cloud: ISO/IEC 42001 – Compliance · 3 November 2026 · primary source
- TechCrunch: Google removes pledge to not use AI for weapons from website · 4 February 2025 · secondary source
- Built In: Google Is Deepening Its AI Ties With the Pentagon · 9 December 2025 · secondary source
- Democracy Now! (Bericht über +972 Magazine, Local Call, The Guardian): Report: Google and Amazon violated their own terms of service in Israel's $1.2 billion Project Nimbus deal · 31 October 2025 · secondary source
- Fortune: Google and Character.AI agree to settle lawsuits over teen suicides · 8 January 2026 · secondary source
- Tech Justice Law Project: Garcia v. Character Technologies, Google, and Character.AI co-founders · 21 May 2025 · secondary source
- TechCrunch: Google says it will sign EU's AI code of practice · 30 July 2025 · secondary source
- Future of Life Institute: AI Safety Index – Winter 2025 · 3 November 2026 · primary source
2 · Safety
Frontier models are tested with disclosed limitations and external testers, but several lawsuits and investigations into harmful answers rule out “strong”.
What is documented
- The Gemini 3 Pro report tests CBRN, cyber, manipulation, ML research and misalignment and openly names the limits of the tests1
- The Gemini 3.1 Pro model card contains automated safety evaluations and external manual red-teaming2
- According to the company, Apollo Research, Vaultis and Dreadnode test externally; UK AISI received early access3
- Since October 2025 a reward programme for AI vulnerabilities has paid up to USD 30,00010
- The Threat Intelligence Group publishes reports on misuse of generative AI and blocks state actors13
The case against
Harmful and false answers
- According to a lawsuit, Gemini fostered emotional dependence and guided a user towards suicide; Google points to crisis notices16
- In two defamation lawsuits, Bard, Gemma and AI Overviews allegedly made false accusations against people and companies; the cases are ongoing6 7
- According to an investigation, AI Overviews gave misleading health information; Google removed them only for individual queries8
Sources for Safety (17)
- Google DeepMind: Gemini 3 Pro – Frontier Safety Framework Report · 3 November 2026 · primary source
- Google DeepMind: Gemini 3.1 Pro – Model Card · 19 February 2026 · primary source
- Google / Google DeepMind: 2026 Responsible AI Progress Report (AI Responsibility Update 2026) · 3 November 2026 · primary source
- Future of Life Institute: AI Safety Index – Winter 2025 · 3 November 2026 · primary source
- TechCrunch: Google pulls Gemma from AI Studio after Senator Blackburn accuses model of defamation · 2 November 2025 · secondary source
- Let's Data Science: Delaware Court Lets Google AI Defamation Case Proceed · 3 November 2026 · secondary source
- Reason – The Volokh Conspiracy: Google Missed Key Deadline in Suit Alleging Google's AI Libeled Business, Court Holds · 12 January 2026 · secondary source
- TechCrunch (report on Guardian research): Google removes AI Overviews for certain medical queries · 11 January 2026 · secondary source
- CNBC: Google pauses Gemini AI image generator after inaccuracies · 22 February 2024 · secondary source
- The Register: Google declares AI bug hunting season open, sets a $30K max reward · 7 October 2025 · secondary source
- Infosecurity Magazine: Gemini Trifecta Highlights Dangers of Indirect Prompt Injection · 30 September 2025 · secondary source
- Google: Gemini Apps Privacy Hub · 26 October 2026 · primary source
- Google Threat Intelligence Group: Adversarial Misuse of Generative AI · 29 January 2025 · primary source
- TechCrunch (report on Common Sense Media): Google Gemini dubbed 'high risk' for kids and teens in new safety assessment · 5 September 2025 · secondary source
- U.S. Federal Trade Commission: FTC Launches Inquiry into AI Chatbots Acting as Companions · 11 September 2025 · primary source
- Semafor: A new lawsuit claims Gemini assisted in suicide · 4 March 2026 · secondary source
- Bloomberg Law: Google Beats Suit Over Data Tracking by Gemini AI Assistant · 8 July 2026 · secondary source
3 · Transparency
Google publishes model cards, test reports and a privacy overview for the Gemini app, but says little about training data.
What is documented
- SynthID Seals text, audio, images and video; the text watermark is open source, and C2PA is used in Pixel 10 and Nano Banana Pro1
- Model cards and test reports for Gemini 3 Pro and 3.1 Pro are published and name intended uses and limitations5 4
- The privacy overview for the Gemini app discloses the use of chats for training and the retention periods10
- The 10-K annual report names AI risks for human rights and privacy6
The case against
Delayed or sparse information
Sources for Transparency (13)
- Google / Google DeepMind: 2026 Responsible AI Progress Report (AI Responsibility Update 2026) · 3 November 2026 · primary source
- Full Fact: Google's SynthID Detector flagging apparently real images · 21 October 2026 · primary source
- Creati.ai (report on The Guardian): Google criticized for hiding AI health disclaimer in AI Overviews · 17 February 2026 · secondary source
- Google DeepMind: Gemini 3 Pro – Frontier Safety Framework Report · 3 November 2026 · primary source
- Google DeepMind: Gemini 3.1 Pro – Model Card · 19 February 2026 · primary source
- Alphabet Inc. / SEC: Alphabet Inc. – Form 10-K for fiscal year 2025 · 5 February 2026 · primary source
- Al Jazeera (Reuters): Conservative activist sues Google over AI-generated statements · 22 October 2025 · secondary source
- Semafor: A new lawsuit claims Gemini assisted in suicide · 4 March 2026 · secondary source
- Google DeepMind: Frontier Safety (Frontier Safety Framework, overview page) · 17 April 2026 · secondary source
- Google: Gemini Apps Privacy Hub · 26 October 2026 · primary source
- TechCrunch: Google says it will sign EU's AI code of practice · 30 July 2025 · secondary source
- Fortune: British lawmakers accuse Google DeepMind of 'breach of trust' over delayed Gemini 2.5 Pro safety report · 29 August 2025 · secondary source
- Autorité de la concurrence: Related rights: the Autorité fines Google €250 million · 20 March 2024 · primary source
4 · Why the harm rule does not apply
The harm rule does not apply: the Autorité fine concerns training data, not harm caused by the use of AI. The lawsuits over Google's own chatbot answers are ongoing without a judgment, and there is no pattern of serious, well-documented incidents.
5 · Controversies (17)
- mediumAccording to a lawsuit filed by a father, Gemini fostered emotional dependence and guided his son towards suicide. · 4 March 2026 · open
- mediumAccording to a Guardian investigation, AI Overviews gave misleading health information; Google removed them for individual queries. · 11 January 2026 · unklar
- mediumAccording to a lawsuit by Wolf River Electric, AI Overviews defamed the company; the case was remanded to state court. · 9 January 2026 · open
- mediumFamily lawsuits over teen suicides after Character.AI chats; Google is a co-defendant because of a licensing deal, and a settlement in principle has been reached. · 8 January 2026 · Vergleich
- mediumThe European Commission is investigating the use of publishers' content for AI Overviews. · 9 December 2025 · open
- mediumGoogle dropped its pledge against AI for weapons and surveillance and then supplied Gemini to the US Defense Department. · 9 December 2025 · open
Further controversies (11)
- mediumAccording to an investigation, clauses in the Project Nimbus contract with Israel are said to undermine Google's own terms of use. · 31 October 2025 · open
- mediumAccording to a lawsuit by Robby Starbuck, Bard and Gemma falsely accused him of serious crimes. · 22 October 2025 · open
- mediumThe Irish Data Protection Commission is investigating whether Google carried out an impact assessment before training PaLM 2. · 12 September 2024 · unklar
- mediumThe Autorité de la concurrence fined Google EUR 250 million, among other things for training Bard on press content. · 20 March 2024 · abgeschlossen
- minorA class action over Gemini features in Gmail and Chat allegedly activated without consent was dismissed. · 7 July 2026 · abgewiesen
- minorFour AI-related shareholder proposals, including one on board oversight of AI, were rejected. · 5 June 2026 · abgewiesen
- minorGemma invented rape allegations against Senator Blackburn; Google removed the model from AI Studio. · 2 November 2025 · unklar
- minorTenable reported prompt-injection flaws in Gemini with a risk of data leakage; fixed according to trade press. · 30 September 2025 · unklar
- minorThe FTC is requiring Alphabet to report on the effects of companion chatbots on children and teenagers. · 11 September 2025 · open
- minorThe Gemini image generator produced historically inaccurate depictions of people; Google paused the feature. · 22 February 2024 · unklar
- minorThe antitrust lawsuit by Penske Media and Chegg over AI Overviews was dismissed; an appeal is pending. · · dismissed
6 · What remains open
- Independent verification of the safety claims, such as UK AI Security Institute reports on pre-release testing
- Incident statistics or a public incident process for AI systems
- Grievance channels with remedy for affected people
- Human rights due diligence for military and government clients
- Court records of the Garcia, Starbuck and Gavalas lawsuits and original reports by the European Commission, the CMA and the Irish Data Protection Commission
- Outcome of the Irish Data Protection Commission's investigation into PaLM 2
7 · How it was assessed
Cut-off date: 3 November 2026 · 43 searches, 40 fetches.
Primary documents read (12)
- Alphabet Form 10-K for 2025 (p. 27, Item 1A, partial)
- Alphabet Proxy Statement DEF 14A 2026 (p. 15, partial)
- Alphabet Form 8-K Voting Results 2026 (p. 36)
- 2026 Responsible AI Progress Report (p. 16)
- Google AI Principles (p. 19)
- Frontier Safety Framework overview (p. 20) and Gemini 3 Pro FSF Report (p. 37)
- Gemini 3.1 Pro Model Card (p. 18)
- Gemini Apps Privacy Hub (p. 35)
- Google Cloud ISO/IEC 42001 (p. 22)
- Autorité de la concurrence, press release on fine (S04)
- FTC, press release on 6(b) study (S28)
- GTIG report: Adversarial Misuse of Generative AI (S32)
Research of 3 November 2026 under Globalance AI Footprint rulebook version 0