Two researchers have explained their departures from Google DeepMind with concerns about military AI and insufficient oversight. The BBC published direct interviews with research engineer Sofia Liguori and AI researcher Andreas Kirsch on October 21. Liguori had announced her final day in her own statement the day before.
Liguori criticizes Google’s growing involvement in military and surveillance systems. She argues that discussion should not focus only on agents exceeding their instructions: deliberate integration into sensitive systems also needs scrutiny. Her assessment of the consequences is explicitly a personal position.
Kirsch told the BBC that he considers Google’s oversight insufficient for safe military and classified deployments, or for developing artificial superintelligence safely in the future. He believes Google has limited visibility and influence over Pentagon use, and says current models are not robust enough for military tasks. These statements are not accompanied by an independent examination of complete contracts.
Google told the BBC it follows a consistent process for public-sector contracts, including compliance with its acceptable-use policies and AI principles. It said employees have channels to raise concerns and workplace rules to prevent retaliation. That response stands alongside the former employees’ assessments.
The BBC also interviewed former OpenAI geopolitics lead Sarah Shoker, who warned that AI can compress decision time until human operators struggle to check large numbers of proposed targets. This is a concern about meaningful human oversight, not technical proof of a particular erroneous strike. The report gives no precise departure date for Kirsch or evidence that he and Liguori coordinated their resignations.