Synopsis
One97 Communications, Razorpay Software and Pine Labs are among the Indian companies that have pushed the San Francisco-based AI developer to let them test Mythos and detect vulnerabilities on their own systems. Their requests came after Anthropic announced a limited roll-out of its latest large-language model, which it considers too dangerous to release more widely.Listen to this article in summarized format
One97 Communications, Razorpay Software and Pine Labs are among the Indian companies that have pushed the San Francisco-based AI developer to let them test Mythos and detect vulnerabilities on their own systems. Their requests came after Anthropic announced a limited roll-out of its latest large-language model, which it considers too dangerous to release more widely.
“We had an urgent call with Anthropic to check when they’re creating a second list of companies that will get access to Mythos,” said Vijay Shekhar Sharma, founder and chief executive of One97.
Sharma said Anthropic representatives asked him what One97 would do with Mythos, and how it could help the company. The questions, he suggested, reflect just how seriously Anthropic is weighing who gets near this technology — and why.
“Is this the beginning of the end?” said Sharma, adding that the anxiety over such models is existential not just for businesses but for financial systems. “A country’s technology networks and financial systems could be infiltrated from any node anywhere. You don’t need to fire missiles to go to war anymore.”
The push among Indian firms to win access to Mythos reflects fears across the world. Regulators, central bankers and executives have been on high alert after it emerged that Mythos can discover cybersecurity vulnerabilities that have gone undetected for years.
Anthropic first stress-tested Mythos internally before raising the alarm and extending access to a select group of a dozen companies, including Amazon Web Services, Apple, and JPMorgan Chase & Co. The AI company is looking to cautiously expand that access through a program it calls Project Glasswing.
The move sent tremors through global financial circles, with US Treasury Secretary Scott Bessent calling the model “a step function change in abilities” — meaning a sharp jump. European Central Bank President Christine Lagarde warned this week about the risk if Mythos fell into the wrong hands.
The question now haunting boardrooms and government ministries alike is whether Mythos could enable the mass looting of bank accounts, paralyse international payment systems or even trigger a full-blown crisis in the global financial system. The Reserve Bank of India didn’t immediately respond to an email on whether the banking regulator or India’s banks and insurers were taking any steps to assess the risks from Mythos.
Security teams are already working overtime at Razorpay, a company whose platform businesses use to collect payments through credit and debit cards, online banking and electronic wallets.
“It’s a race against time for startups like us,” said Razorpay’s cofounder and chief executive officer Harshil Mathur. “We’ve asked for Mythos access as we want to test the weaknesses on our platform and strengthen our defenses.”
Mathur said that in the startup groups he’s part of, Mythos has been the hot topic of discussion for the past 10 days. Anthropic may expand the roll-out with stringent contracts on limited use to test companies’ own infrastructure, disallowing commercial use, he said.
India is home to millions of engineers who write code for Wall Street banks, insurers and credit card giants. The country has become the second-largest market for Anthropic’s Claude model, with coders mainly using it for building apps, debugging software and modernizing IT systems.
“Regulators will push for more stringent security norms owing to the threat of increased attacks,” said Pine Labs’ chief executive officer Amrish Rau. “Security can’t be a compliance checkbox anymore.”
