Visa open

Visa has open-sourced part of its AI-powered cyber defence system as it prepares for increasingly sophisticated attacks.

The payments giant is among a growing number of organisations expressing concern the risks posed by advanced AI, including autonomous cyberattacks and the longer-term threat of quantum computing.

Visa processes approximately one billion payments a day, worth around $15 trillion annually, making it a critical component of the global financial system.

Its president of technology, Rajat Taneja, told Reuters that the company had released part of its cyber defence system as open-source software following the discovery of vulnerabilities by Anthropic's Mythos AI model earlier this year.

Taneja described the weaknesses as "humbling", warning that the incident involving AI agents escaping a testing environment and attacking the Hugging Face platform could be an early indication of more serious threats to come.

"It's hard to predict how bad it could get, but we have seen the trailer," he said.

Taneja said future cyberattacks could involve autonomous systems that continuously learn and adapt without human intervention, potentially giving attackers a significant advantage over conventional security measures.

The warning carries particular significance for payment companies, where maintaining customer trust is essential to their operations.

Visa has also begun allowing some AI agents to make payments on its network, although Taneja declined to disclose how many transactions were being conducted this way.

Industry estimates suggest AI agents could handle approximately a third of online commerce by 2030, representing close to $3.1 trillion in transactions.

Taneja also outlined the potential threat from quantum computing, warning that sufficiently powerful machines could eventually use Shor's algorithm to break encryption standards underpinning global commerce.

He said businesses would need to continue developing ways to mitigate the risk.

AI race is outpacing safety measures

As Visa makes its AI-powered cyber defences open source, current and former researchers at OpenAI and Google DeepMind have warned that technology companies are moving too quickly to develop self-improving AI systems, potentially putting humanity at risk .

In video testimonials gathered by AI safety nonprofit Palisade Research, researchers stressed that their concerns existential risks were sincere and not intended to generate publicity.

The project, called frominside.ai , aims to bring researchers' concerns directly to the public rather than confining the debate to social media and specialist communities.

"The risk is ramping up pretty fast," Geoffrey Irving, co-founder of AI nonprofit Resolution told Reuters.

"It's on me and the rest of the field to be direct," said Irving , who has previously worked at both OpenAI and DeepMind.

In one video, DeepMind research scientist Neel Nanda said that AI posed at least a 10% risk of causing human extinction, describing the probability as "ridiculously high".

Juan Felipe Ceron Uribe, an AI alignment research engineer at OpenAI, said developers should take much greater care when building advanced systems.

Researchers are particularly concerned recursive self-improvement, a process through which AI systems could continuously acquire new capabilities with little or no human involvement.

They argue that future generations of models could advance faster than existing safety mechanisms, leaving society unprepared for the consequences.

Anthropic chief executive Dario Amodei published an essay this month calling on the industry to slow down and "pace the frontier", a position quickly supported by OpenAI chief executive Sam Altman .

Both companies have released new models this month as they compete for customers, although OpenAI said this week that it had delayed the launch of an even more powerful system.