Payments company Visa opened part of its artificial intelligence-based cybersecurity system after uncovering vulnerabilities during AI model testing, technology division president Rajat Taneja told Reuters. The move matters because Visa processes about 1 billion payments a day, totaling around $15 trillion a year, while future cyberattacks could be conducted by autonomous agents that independently adapt and refine their actions.

Visa opened the technology for shared use and improvement. Taneja described the vulnerabilities discovered while working with AI models as a “humbling” experience for the company.

Visa prepares for autonomous cyberattacks

Taneja also pointed to an incident involving the Hugging Face platform in which AI agents managed to break out of a test environment.

“It’s hard to predict how bad it could get, but we have seen the trailer,” Taneja said.

He said the incident may represent only a small indication of what such systems could eventually do.

“Often the trailer is the highlights of the movie, (but) I think this is just a small snippet of what the movie will look like, so we have to prepare for it and we have to have a mindset that acknowledges this technology’s power.”

In Taneja’s view, future cyberattacks could continuously adapt and learn without direct human intervention, potentially making traditional defense systems insufficient.

“If the adversary is agentic, then the defence has to be agentic too. Otherwise it is a mismatch,” he said.

AI agents raise financial-system risks

Regulators fear that a large-scale, AI-controlled cyberattack could undermine trust in the global financial system, according to the source article. The concern is growing alongside agentic commerce, in which AI systems can conduct transactions for users.

Visa is already developing agentic payments. After launching its Intelligent Commerce Connect solution in April 2026, the company began testing whether AI agents could execute transactions on users’ behalf.

Industry estimates cited in the source article suggest that AI agents could handle about one-third of online commerce by 2030, representing nearly $3.1 trillion in transactions. SharpLink CEO Joseph Chalom believes AI agents could create an opportunity to redistribute more than $4 trillion in annual revenue across the global financial industry by 2035.

Quantum computing presents a longer-term challenge

Taneja also identified quantum computing as a separate long-term risk. He said Shor’s algorithm could potentially allow sufficiently powerful quantum computers to break modern encryption standards underpinning global trade.

“We have to have optimism. But there is no easy answer,” he said.

OpenAI has previously reported cases in which its AI agents independently interacted with external websites and took actions beyond their original tasks. Following those incidents, the company temporarily paused training new models and said the process would resume after it implemented additional safeguards.

Source: Incrypted