Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

Aton launches 'digital power of attorney' that caps AI transfers at 100,000 won

Aton, a Korean financial-technology company, has launched a 'digital power of attorney' product for the financial sector, according to a Money Today report, with AI-initiated transfers limited to 100,000 won. The launch addresses the authorisation layer that banks and brokers need before letting AI agents act on customer accounts.

Published

Aton, a Korean financial-technology company, has launched a 'digital power of attorney' product aimed at financial institutions, according to a Money Today report published on November 11. The headline states the central constraint bluntly: AI is allowed to transfer only up to 100,000 won.

The product addresses a question that appears as soon as AI assistants start handling money on a customer's behalf: how does an institution verify what an agent is authorised to do, and how far it may go? A 100,000-won transfer ceiling turns that abstract question into a concrete, checkable limit.

Paper powers of attorney and counter verification have traditionally carried this role, but they are slow and awkward to fit into automated flows. A digital authorisation record moves the delegation online, giving a bank or broker a machine-readable basis for checking whether an AI-initiated action sits inside the scope the customer granted, rather than leaning entirely on manual counter checks.

Writing an amount limit into the authorisation is a pragmatic way to let agentic AI touch real money: cap the exposure first, then widen it as behaviour proves reliable. Compared with handing an agent full account access, a bounded mandate is far easier to defend to compliance teams and to explain to customers.

It should be noted that public reporting so far discloses only the amount ceiling. Aton's partner institutions, a rollout timetable, and how customers would grant or revoke authorisation have not been made public, which suggests the product is still early and its practical reach is unclear.

For the industry, the signal matters more than the product itself: before AI agents can genuinely act for people, the plumbing of authorisation, limits and liability has to exist. Whoever turns that credential layer into something financial institutions will actually integrate holds the entry point for agents in financial services.

Why it matters

If digital mandates of this kind mature, AI agents could move into transfers, wealth management and insurance far faster. The open question is who bears liability when a bounded agent exceeds its mandate, a point the current reporting does not yet answer.

AtonAI AgentFintech
Back to realtime news

Nearby Updates

All

11/12, 07:47

Bank of Montreal rolls out AI Agent Assist, pushing agents into bank workflows

Simply Wall St reported that Bank of Montreal (TSX:BMO) has rolled out a tool called AI Agent Assist, presenting it alongside a narrative that the stock looks modestly undervalued. The headline and summary give no detail on capabilities, business lines covered or rollout scope, leaving the specifics to official disclosure.

11/12, 07:58

Animoca puts Yat Siu in front of its AI agent push

The Mercury reports that Animoca is leaning on co-founder Yat Siu to lead its AI agent push, in a one-line item that gives no detail on title, team or roadmap. The signal is still clear: a company known for Web3 and digital-asset bets is moving agents toward the center of its strategy.

11/12, 07:28

Nvidia shares slide as board reportedly prepares to remove CEO Jensen Huang

TechCrunch reported on November 11 that Nvidia's stock fell as the company's board reportedly prepared to remove founder and CEO Jensen Huang, possibly as soon as Wednesday evening. The report is hedged as unconfirmed, leaving the timing of any board vote and the shape of any succession plan unknown.

11/12, 06:52

Claude Mythos moves into Cisco and IBM security platforms

Anthropic's Claude Mythos is expanding into Cisco's and IBM's security platforms, according to a Security Boulevard report. The signal is less about model capability than about distribution: the model lands inside the platforms security teams already run instead of arriving as a separate endpoint.