AI firms like Anthropic can slow down, but can't ask others while they keep building: Zoho's Sridhar Vembu

Artificial intelligence companies that believe their technology could pose a danger can slow down or stop their own development, Zoho cofounder and chief scientist Sridhar Vembu said, questioning the commercial motives behind calls for others to hold back.

Vembu’s comments come less than two weeks after Anthropic CEO Dario Amodei called for the AI industry to slow the pace at which it develops increasingly capable models. On September 12, Amodei argued that AI capabilities were advancing faster than the safeguards needed to manage them. OpenAI CEO Sam Altman and xAI chief Elon Musk subsequently backed the call.

Yet the two AI labs have continued to push out new models. On Tuesday, Anthropic launched Claude Opus 5.5, while OpenAI unveiled GPT-6 Sol and GPT-6 Luna.


Also Read: Anthropic CEO Dario Amodei calls for slower AI progress, warns of 'catastrophic' risks

Speaking to ET AI, Vembu said he did not dismiss concerns about AI’s potential dangers, but remained unconvinced that the technology posed an imminent threat of human extinction.

Referring specifically to Anthropic, Vembu said the company could stop development itself if it wanted to. “They can slow it down themselves. They don’t need anybody’s permission,” he said.

Vembu questioned the position of companies that, in his view, wanted others to stop while continuing their own investments. “What they are saying is, let the others stop doing it. We will continue investing in it. That’s not acceptable,” he said.

“For example, I am working on a piece of technology I believe is dangerous; I have the freedom to stop working on it,” Vembu said.

However, Vembu said warnings from people without a commercial interest also deserved consideration. “I don’t want to dismiss it all as hype. But at the same time, I don’t particularly believe that either,” he said.

Vembu said AI risks should be taken seriously because many smart people had raised concerns, but he was not convinced the danger was as great as they suggested.

“In all of these, we have to be scientists. We have to go with the evidence,” he said.

Productivity with safety

While questioning warnings that AI could threaten humanity, Vembu said he spends about 70% of his time making AI-generated software safer and more reliable.

Vembu compared AI coding tools to powerful machine looms that could produce large amounts of code, but said the speed of production raised questions about accuracy, reliability and security.

He said conventional testing might not uncover every problem.

“Even though it passes all the tests, there could still be a piece of malware sitting there,” he said.

At Zoho, Vembu said, AI was already being used extensively in programming, but human developers remained responsible for the software they released.

“That is why we are not simply chasing productivity alone. We want productivity with safety,” he said.

Also Read: Zoho brings AI to everyday work with Zia Chat, adds seven vertical SaaS offerings

Vembu compared the challenge to running a high-speed train: “A fast train also requires reliable tracks, signalling and safeguards.” Similarly, he said, AI systems needed supporting tools that could cross-check their output before productivity could safely increase.

Vembu said questions about who would be responsible for serious AI mistakes would have to be addressed through laws and regulations, as well as stronger safeguards built into the technology itself.

Keeping humans accountable

Vembu said the need for human responsibility also applied to decisions about employees. Asked whether AI agents could conduct the first round of appraisals, he said: “We wouldn’t do it.”

He said human evaluators could seek help from AI, but must remain in charge of the assessment.

Vembu said Zoho’s use of AI in customer support had not displaced its support staff, adding that the company was hiring more people. Vembu also warned that junior employees could see a short-term productivity increase from using AI too quickly without learning in the process.

He said organisations needed guidelines for appropriate AI use to ensure employees continued developing their skills, stressing that junior staff must keep learning as they use these tools.

“We are still in the learning phase in all this,” he added.