Industry weighs AI safety against innovation

By Milan Rojan

The rapid advancement of artificial intelligence (AI) is creating new opportunities for enterprises, but increasingly stringent safety controls on foundation models could limit innovation and real-world deployment, according to Staqu Co-founder and CEO Atul Rai.

His comments follow the recent unveiling of Claude Mythos, Anthropic’s latest AI model, which has attracted industry attention for combining enhanced capabilities with an extensive safety architecture designed to mitigate misuse and ensure responsible deployment.

As enterprises accelerate investments in AI-driven automation, analytics and cybersecurity, questions are emerging around the balance between model safety and performance. While safeguards are widely viewed as essential for preventing harmful outcomes, some industry leaders argue that overly restrictive controls may hinder organisations seeking to build advanced AI applications tailored to complex business requirements.

“AI delivers measurable business impact by helping organizations either generate revenue or reduce costs,” said Atul Rai, Co-founder and CEO of Staqu.

Rai noted that the debate around enterprise AI is increasingly shifting beyond model capability to the constraints surrounding deployment and customisation.

“Yet today, one of the most pressing tensions in enterprise AI is not capability but constraint. As public-facing AI models continue to mature, their safety guardrails, while well-intentioned, are increasingly becoming a bottleneck for organizations looking to build and deploy more sophisticated AI solutions,” he said.

The discussion has gained momentum as businesses seek greater flexibility in deploying AI across areas such as fraud detection, cyber-defence, risk management and operational automation. Financial institutions in particular are exploring how advanced AI systems can enhance decision-making while meeting regulatory and security requirements.

Referring to Anthropic’s latest model, Rai said the launch highlighted a broader challenge facing the industry.

“While it represents Anthropic’s most powerful model to date, its extensive safety architecture raises a critical question for the industry: at what point do guardrails designed to ensure responsible deployment begin to limit the very innovation they are meant to support?” he said.

According to Rai, the issue is especially relevant for India’s ambitions to strengthen AI-powered cybersecurity capabilities. He argued that unlocking the full potential of advanced AI systems will require greater autonomy and adaptability without compromising safety standards.

“This challenge is particularly consequential for India’s ambitions in AI-powered cyber-defense,” said Rai. “The industry must move beyond treating safety and capability as opposing forces and instead build frameworks where robust guardrails and real-world performance can genuinely coexist.”

The debate reflects a growing industry focus on developing AI governance frameworks that balance innovation, security and compliance as organisations move from experimentation to large-scale deployment.