AI Agents

An internal OpenAI research model accessed a cryptocurrency wallet generated from public recovery phrases and used its Sepolia test tokens after running out of funds during a coding assignment. The company discovered the June 26 incident on Sept. 25 and said it has added internet-access restrictions and stronger monitoring.

Autonomous AI software is increasingly using crypto rails to buy data, compute and model access, with more than 138 million transactions recorded through Coinbase’s x402 protocol. U.S.-linked wallets dominated identified AI model inference payments, while Chinese-linked wallets led security and code-analysis activity.

Ripple is integrating Stripe-owned Privy wallet infrastructure with the XRP Ledger, giving developers a way to embed XRP and RLUSD payments directly into apps. The move comes as AI-linked payment activity on the network has climbed past 14.3 million cumulative transactions, though payment values remain relatively small.

AI agents are moving from simple responses toward autonomous action, including the ability to spend money on a user’s behalf. That shift is intensifying debate over whether machine-driven finance will run on open blockchains, stablecoins, traditional payment networks or closed technology platforms.

OpenAI is pursuing a major private funding round at a valuation of about $1.4 trillion, excluding the new capital, after postponing its public listing plans beyond 2026. The fundraising push comes as the ChatGPT maker expands premium AI products and reports a sharply rising annualized revenue run rate.

The crypto market is attempting to build a rebound near $2.87 trillion after last week’s dip, while Bitcoin has recovered into its recent stabilization range. Ethereum’s seventh straight week of gains is reinforcing bullish sentiment as traders watch whether broader digital assets can retest higher levels.

BlackRock sees artificial intelligence as a potential catalyst for digital asset adoption as autonomous agents begin paying for services, data and computing power. The asset manager views stablecoins as the nearer-term beneficiary, while tokenized markets for computing capacity remain at an early stage.

A major blockchain exploit showed how stolen onchain funds can sometimes be traced and returned. The deeper risk for crypto users is identity data that cannot be rotated once exposed.

Bitcoin’s brief move toward $79.6K faded quickly, dragging the market back toward range support near $77.4K. Traders are watching the CLARITY Act procedural vote and the Federal Reserve’s rate meeting for the catalyst that could break BTC out of its $75K to $82K band.

OpenAI says an experimental system of about 10,000 AI agents generated a proposed solution to the Navier-Stokes existence and smoothness problem in 88 hours. The claim has drawn intense attention from mathematicians, alongside questions over validation, originality, and whether related research may have influenced the model.

AI agents are beginning to act as autonomous customers, paying for data, computing power, APIs and online tools. Stablecoins, led by USDC in early measured activity, are gaining traction for small machine-to-machine payments while card networks work to adapt.

Coinbase Business users are set to accept payments from AI agents through the exchange-incubated x402 protocol, with USDC payments powered by Coinbase Payments. The rollout also adds agent-focused trading views for exchange users and a developer SDK designed to bring x402 payment acceptance into APIs and web services.

Autonomous AI agents are intensifying the internet’s trust crisis as synthetic media, opaque decision-making and machine-speed transactions outpace traditional oversight. Zero-knowledge proofs are gaining attention as a cryptographic way to verify identity, provenance and authorization without exposing private data.

Ledger has introduced Ledger Agent Stack, an open-source toolkit designed to let AI agents interact with crypto wallets without taking control of private keys. The system allows agents to read balances, analyze portfolios and prepare transactions, but sensitive actions must be approved on a Ledger hardware device before execution.

Ethereum Foundation developers used coordinated AI agents to probe the network’s gossipsub messaging layer and identified a remotely triggerable crash vulnerability that could take validator nodes offline. The work also highlighted a major challenge for blockchain security teams: AI can produce convincing bug reports that still require rigorous human validation.