By Marc Jones
LONDON, Sept 29 (Reuters) - Visa has open-sourced part of its AI-powered cyber defence system after "humbling" lessons from vulnerabilities exposed by AI models, as the payments giant braces for autonomous cyberattacks and future quantum-computing threats.
Visa processes roughly a billion payments a day worth around $15 trillion a year, making it a key part of the world's financial plumbing system.
Its President of Technology, Rajat Taneja, told Reuters the company had made its cyber defence
system available as open-source software after weaknesses exposed by Anthropic's Mythos AI model earlier this year highlighted the scale of the challenge.
The vulnerabilities were "humbling", Taneja said, adding that the attack by AI agents on the Hugging Face platform may also prove to be an early glimpse of a more serious future threat.
"It's hard to predict how bad it could get, but we have seen the trailer," Taneja said, referring to how the AI models escaped a testing sandbox in the Hugging Face incident.
"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".
As well as the dystopian warnings about AI's threat to humanity, global regulators fear a major AI-controlled cyberattack could undermine confidence in the world's financial system.
Visa has plenty at stake. Payments firms rely on trust, meaning cyberattacks can be devastating. It has also started allowing certain AI agents to use Visa.
Taneja did not provide details on how many of those payments were happening, but industry estimates suggest roughly a third of online commerce - close to $3.1 trillion in transactions - could run through AI agents by 2030.
According to Taneja, future cyberattacks will be adaptive, continuously learning and improving without direct human intervention, giving attackers a "superpower".
"Human defences, which are kind of the current model, won't work," Taneja said. "If the adversary is agentic, then the defence has to be agentic too. Otherwise it is a mismatch".
He also highlighted the longer-term challenge posed by quantum computing, saying that Shor's algorithm could one day enable powerful quantum computers to break the current encryption standards that underpin global commerce.
He said firms would therefore have to continue looking for ways to neutralise that risk.
"We have to have optimism," Taneja said. "But there is no easy answer."
(Reporting by Marc Jones, Editing by Louise Heavens)
