Concerns over the behaviour of OpenAI agents have sparked a wider debate about AI safety and whether the development of advanced artificial intelligence should be slowed down. However, NITI Aayog Distinguished Fellow Debjani Ghosh believes the recent incidents should not be viewed as proof that AI systems have become uncontrollable.
Ghosh argued that the behaviour of AI agents reflects the objectives, boundaries and reward systems created by humans. According to her, the focus should therefore be on improving human capabilities, strengthening AI safeguards and establishing effective guardrails for the development of increasingly powerful AI systems.
Debjani Ghosh Calls for Better AI Safety Measures
In a post on X, Ghosh said that a temporary and carefully planned pause in AI development could make sense, but only if the additional time is used productively.
She suggested that such a pause should focus on building stronger safety mechanisms and improving the ability of humans to develop, deploy and manage AI responsibly.
Ghosh cautioned against framing the discussion simply as a call to “slow down AI”. In her view, the objective should instead be to use any pause to improve human expertise and establish meaningful safeguards around advanced AI development.
‘AI Did What Humans Trained It to Do’
Ghosh also challenged the argument that recent OpenAI agent behaviour demonstrates that artificial intelligence is independently moving beyond human control.
Her view is that the systems followed the objectives they were given. If developers create inadequate boundaries or reward models for achieving a particular outcome, an AI system may find unexpected ways to accomplish that objective.
She described this as a failure in how the system was specified, rather than evidence of AI spontaneously developing behaviour outside its intended framework.
In other words, the problem may not be that AI has suddenly become uncontrollable, but that humans need to become better at defining what AI systems should and should not be allowed to do.
Why a Pause in AI Development May Not Be Enough
Ghosh said she could support a planned pause in AI capability development if there is a clear strategy behind it.
Simply slowing development without simultaneously investing in AI safety, governance and human expertise, she argued, could amount to little more than postponing the problem.
For a pause to deliver meaningful safety improvements, resources would need to be directed towards developing better evaluation systems, stronger safeguards and the expertise required to build and manage advanced AI.
Global Debate Over Slowing AI Development
The discussion around whether AI development should be slowed has intensified across the global technology industry.
Anthropic CEO Dario Amodei has previously called for a slower pace of AI development, citing concerns about the risks associated with increasingly capable AI systems and the possibility of humans losing control over advanced models.
Several other prominent technology leaders have also weighed in on the issue, including OpenAI CEO Sam Altman, Tesla CEO Elon Musk and Google DeepMind CEO Demis Hassabis.
Their views have added momentum to the broader debate over how the industry can continue developing powerful AI while ensuring that appropriate safety measures keep pace with those capabilities.
Microsoft CEO Satya Nadella Backs ‘Deliberate Pacing’
Microsoft CEO Satya Nadella has also expressed support for greater caution in the development of advanced AI.
Nadella has argued that the pursuit of superintelligence should remain grounded in a fundamental principle: AI must continue to benefit humanity and remain under meaningful human control.
He has also supported research and what he described as “deliberate pacing”, with AI alignment treated as an important design objective rather than something addressed only after a system has been developed.
One of the approaches he has pointed to is the use of embedded evaluators, which could help assess AI behaviour and improve safety during the development and operation of advanced systems.
AI Benefits Should Be Widely Distributed
Nadella has also highlighted the importance of ensuring that the benefits of advanced AI are not concentrated among a small number of companies or organisations.
According to his position, access to powerful AI capabilities should extend across countries, communities and businesses rather than being controlled by only a handful of major technology players.
He has also called for broader participation in the AI ecosystem, including greater involvement from academic institutions.
What Debjani Ghosh’s Comments Mean for the AI Debate
Ghosh’s comments shift the focus of the AI safety debate from simply asking whether AI development should stop to examining how humans design, train and govern AI systems.
The central argument is that increasingly capable AI requires equally capable human oversight. Better-defined objectives, carefully designed reward systems, robust evaluations and effective guardrails could all play a role in reducing unintended behaviour.
The debate is therefore no longer limited to whether AI development should be faster or slower. It is increasingly about finding the right balance between innovation, safety, human oversight and responsible AI governance.
As AI systems become more capable, the challenge for governments, researchers and technology companies will be to ensure that human expertise and safety frameworks develop alongside the technology.