Former researchers question oversight of AI used in classified military operations as Google defends its contracting policies

Google is facing renewed internal criticism after employees at its artificial intelligence division, Google DeepMind, resigned over concerns about the company’s growing involvement in military AI contracts with the United States and Israel.

According to a BBC investigation published on October 21, 2026, London-based research engineer Sofia Liguori and AI researcher Andreas Kirsch are among those who have left the company amid concerns about the military applications of its technology and whether adequate safeguards are in place.

Liguori, who worked at Google for three years, said she could no longer support the company’s direction as its involvement in military operations expanded. Kirsch separately questioned whether the company’s governance arrangements were sufficient to ensure that its AI systems were deployed safely in classified environments.

Their departures have renewed debate over the responsibilities of technology companies supplying artificial intelligence to defense agencies, particularly when the systems may influence intelligence assessments, operational planning and battlefield decisions.

Why Are Google Employees Resigning?

Liguori told the BBC that the company’s increasing involvement in military AI had changed how she viewed her employment.

She said she felt ashamed to work at Google after realizing the company was pushing further into AI systems for military operations. She also cited concerns about its work with the US and Israeli militaries.

Her objections reflect a wider debate within the technology industry over whether developing powerful AI systems for military customers is compatible with commitments to human rights and responsible innovation.

Kirsch, another former Google DeepMind researcher, raised concerns about the company's ability to supervise how its technology is used once it enters classified military environments.

He argued that Google appeared to have limited visibility into the Pentagon's use of its AI systems and questioned whether the company could effectively control or veto particular applications.

The BBC also spoke to current employees who described anxiety among some staff about the company's changing approach to AI ethics. Some workers reportedly fear that speaking publicly about their concerns could put their careers at risk.

Google, however, said it maintains channels through which employees can raise concerns and has workplace policies designed to prevent retaliation.

Google's Defense Contracts Under Scrutiny

Google is among several major technology companies that have entered agreements to provide AI capabilities for US military work, including classified operations.

Other companies involved in military AI arrangements include OpenAI, Microsoft, Amazon, Oracle, Nvidia and SpaceX, according to the BBC report.

The US Department of Defense has increasingly prioritized AI adoption as part of its military modernization efforts.

Google's agreement allows its AI technology to be used for lawful government purposes, subject to specified restrictions. The contract states that its models should not be used for domestic mass surveillance or autonomous weapons, including target selection, without appropriate human oversight and control.

Nevertheless, critics argue that restrictions written into contracts do not automatically guarantee that AI systems will be used safely in complex operational environments.

Google told the BBC that it follows established procedures when reviewing public-sector contracts and requires compliance with its acceptable-use policies and AI principles.

The company also emphasized that employees have dedicated channels for raising concerns about business ethics and misconduct.

Concerns About AI on the Battlefield

One of the central issues in the dispute is whether current AI systems are sufficiently reliable for military applications where mistakes can have serious consequences.

AI tools can process large volumes of intelligence and information rapidly. In military settings, they may help personnel organize intelligence, assess potential threats, plan operations and identify possible targets.

However, speed does not necessarily guarantee accuracy.

Former employees and experts interviewed by the BBC warned that AI-generated recommendations could be difficult to review adequately when military personnel must make decisions under intense pressure and within short time frames.

Sarah Shoker, formerly head of OpenAI's Geopolitics team, warned that AI can compress the time available for operators to make decisions. She questioned whether human personnel could meaningfully examine the outputs of systems processing hundreds of potential targets in a limited period.

The concerns highlight an important distinction: keeping a human formally involved in a decision does not necessarily ensure meaningful oversight if that person lacks sufficient time, information or technical understanding to evaluate the system's recommendations.

Critics say effective safeguards must therefore extend beyond contractual language to include rigorous testing, operational monitoring, clear accountability and the ability of human operators to reject automated recommendations.

The resigning researchers' concerns do not, by themselves, establish that Google's AI has caused a particular military harm. They do, however, raise questions about the safeguards governing its use in high-risk environments.

Project Nimbus and the Israel Controversy

Google's military-related activities have faced criticism before.

In 2024, employees protested the company's involvement in Project Nimbus, a joint $1.2 billion contract with Amazon providing cloud computing and technology services to the Israeli government and military.

At the time, Google said Project Nimbus was not directed at highly sensitive, classified or military workloads relevant to weapons or intelligence services.

The agreement nevertheless became a focal point for employee concerns about the potential use of commercial technology in military operations.

The latest resignations have renewed scrutiny of the relationship between technology providers and military customers, particularly as AI capabilities become more sophisticated and integrated into defense infrastructure.

The broader debate also encompasses allegations about AI-assisted targeting in Gaza. Such claims have been contested, and the specific role of individual technology providers in particular operations must be distinguished from the wider use of AI-enabled systems by military organizations.

What the Dispute Means for the Future of AI

The controversy presents a challenge for technology companies seeking to expand their government business while maintaining public trust in their safety and ethical commitments.

For AI developers, the questions extend beyond whether a model works as intended during testing. They include whether its performance remains reliable in real-world conditions, whether users understand its limitations and whether a supplier can monitor how its technology is deployed after a contract is signed.

Employee departures may also affect recruitment and workplace culture, particularly if researchers believe their concerns about the social consequences of AI are not being adequately addressed.

However, the scale of any longer-term impact on Google's research operations remains uncertain.

The dispute also reflects a wider policy challenge for governments: how to benefit from AI's potential military applications while establishing meaningful controls over systems that may influence consequential decisions.

As AI becomes more deeply embedded in defense operations, questions about human responsibility, transparency, testing and accountability are likely to remain central to discussions about its appropriate use.

Note: The resignations at Google DeepMind underscore a growing tension in the AI industry between military innovation and responsible technology development. The central question is not simply whether AI can make military operations faster, but whether the systems can be tested, supervised and held accountable when their recommendations may have life-and-death consequences.

Source: BBC News