AI agents making purchases for humans, spending Rs 9,480 per transaction: Stripe

Stripe says AI agents are increasingly making purchases on behalf of humans, with agent-led transactions through its Link payment service rising 38-fold in September. However, reports of AI systems accessing confidential company data have raised concerns about the risks of autonomous technology.

In Short

• AI agent transactions on Stripe Link rose 38-fold in September
• Retail led agent-led purchases, with average spending at around $100
• India's CERT-In warns of AI agents accessing confidential company data

AI agents are beginning to spend money on behalf of humans, handling purchases through payment services without requiring people to complete every step themselves. Stripe's Link payment service recorded a sharp rise in such transactions in September, with the average purchase amount standing at around $100 (approximately Rs 9,480).

According to data shared by Stripe with Upstarts Media, agent-led transactions through Link increased 38-fold in September, while payment volume grew 20 times. Retail emerged as the leading category for both transactions and spending, suggesting that shopping is becoming an early use case for AI agents. These agents can carry out purchases on behalf of users through payment tools integrated into their services. Instead of manually navigating checkout pages, users can rely on an AI agent to handle transactions through supported payment systems.

Stripe's Link currently powers around 200 AI agents, according to Dan Hill, head of consumer product at Stripe, who oversees the service. These include shopping and personal assistant tools such as Instinct, Muse and Grok Bot, alongside work-focused agents like Skydive and Town. The number of agents had more than doubled since Stripe collected its September data, when approximately 70 agents were signed up.

Link has around 300 million users across businesses including Airbnb, Anthropologie and Uber. AI agents still account for a relatively small share of its overall activity, but their use is growing rapidly.

"Agentic commerce is here, people are using it, it's going to keep growing," Hill told Upstarts Media, adding that the payment volume was becoming "very material".

The development points to a new role for AI assistants, which are moving beyond answering questions and generating content to carrying out tasks that involve real money. However, the growing use of autonomous agents also raises questions about the permissions they receive, the transactions they can initiate and the safeguards needed to prevent unauthorised activity.

AI agents also raise cybersecurity concerns

The risks extend beyond payments. India's cybersecurity agency, the Indian Computer Emergency Response Team (CERT-In), recently disclosed that AI models under testing had accessed confidential records and entered live systems belonging to Indian organisations without being detected.

According to Reuters, AI model providers voluntarily reported that their agents had exceeded the permissions assigned to them during testing and accessed company data, including business records. CERT-In did not identify the AI providers or the affected organisations. The agency said none of the affected organisations detected the intrusions themselves. The incidents were reported by the AI companies, bringing India into a wider international debate over the risks posed by AI models with advanced hacking capabilities.

OpenAI has previously disclosed that its models went beyond their assigned tasks during testing and breached a startup. Anthropic has disclosed cases in which its AI systems accessed outside computer networks during testing, adding to concerns about how these models behave when operating independently.

CERT-In separately warned that attackers were using freely available AI tools to target computer systems for financial gain. According to the agency, AI tools had helped compress attacks that once took days to progress from initial access to damage into just a few hours. Organisations that failed to completely remove attackers were also being targeted again within days.

- Ends