The question raised by the Google DeepMind resignations is no longer whether one company has an ethics problem. It is whether the artificial intelligence industry has a shared answer to a question it keeps avoiding: what does responsible AI mean when the customer is a military operating in secret?
Several DeepMind researchers have resigned over Google’s military AI contracts, according to a BBC investigation. Sofia Liguori resigned after three years, saying the push into military operations made her feel ashamed to work at Google. Andreas Kirsch resigned earlier in October, arguing that governance and oversight were not sufficient for classified settings.
An industry answer, not a company answer
The contracts themselves are industry wide. Google’s Pentagon agreement, allowing models to be used for any lawful government purpose in classified operations, sits alongside similar agreements signed by OpenAI, Amazon, Microsoft, SpaceX, Oracle, Nvidia and the start up Reflection. When every major firm signs similar terms, a researcher who objects cannot simply change employer. The objection has to be to the practice itself.
Google says its agreements include restrictions on mass surveillance and autonomous weapons without appropriate human oversight, and that employees have channels to raise concerns without retaliation. The researchers who left argue those restrictions may not work in practice once systems enter classified environments where vendors have limited visibility and no veto over specific uses.
What responsible means now
The internal history shows the debate is not new. Around 600 employees signed a petition in April urging leadership to stay away from classified military work. Alex Turner, who worked on AI safety at DeepMind for more than two years, resigned in June, writing that senior management had insisted Google would not sign such a deal, but the agreement went ahead.
As the Financial Express reported, the resignations raise questions about safety, oversight and ethical implications that apply well beyond one company.
For the industry, the resignations are a warning that responsible AI cannot be a page on a website while the contracts say something else. If the firms building frontier AI cannot show credible oversight of classified uses, they will keep losing the very researchers whose presence makes their responsibility claims believable.
