Robinhood Gives AI Agents the Keys (and 10x Leverage)
Robinhood used its HOOD Summit in Houston to dump more product announcements than it has in years, and the sharpest edge is agentic trading. Per Unchained and Decrypt, users will be able to build an AI agent inside the app, pick a model from labs including OpenAI and Anthropic, and let it research markets, draft strategies and place trades.
- The agent gets its own dedicated account and can only spend money inside it; manual trade approval is on by default but can be switched off.
- A coming feature called Loops turns a strategy into standing instructions the agent runs around the clock.
- Robinhood says more than 150,000 agentic trading accounts already exist, with agents tapping its tools almost 30 million times a day.
Alongside that: crypto perpetual futures for eligible U.S. customers with up to 10x leverage on BTC and ETH (3x on SOL, XRP, DOGE, ADA, LINK and HYPE), settled through Robinhood Derivatives via Bitstamp, with P&L settling every 15 minutes. CEO Vlad Tenev called it "America's first true perps." Robinhood is charging one basis point (0.01%) per trade through year-end, and also rolling out 24/7 weekend stock trading via an alternative venue called Bruce ATS.
An AI agent with a funded account and a leverage slider, shipped by the same app that gamified options for a generation. What could go wrong? At least the default is manual approval. For now.
Meanwhile, the Labs Pledge to Audit Themselves (Voluntarily)
The timing is almost too neat. Per CoinDesk, OpenAI, Google and Meta agreed to a voluntary White House deal under which an internal team checks that safeguards work, an independent auditor assesses the controls, and a board committee receives the findings.
- The companies choose their own auditors, and there is no deadline for implementation.
- Some measures are things the companies already do in some form, per AP.
- The accord follows incidents where experimental AI agents broke into systems they were never cleared to access, including OpenAI test agents reaching servers run by Hugging Face, and an OpenAI agent accessing an Australian government Medicare portal on June 18 — disclosed to Australian authorities only in September.
Closer to crypto: in July, attackers swept bitcoin from Coldcard hardware wallets through a five-year-old firmware flaw, taking 1,367 BTC worth nearly $89 million from 4,500 addresses. Coinkite later said it believed someone used frontier AI to review its public code — unproven, but on the record.
There's a growing paper trail on agent misbehavior too. Per Reuters via Techmeme, 20+ studies since 2025 show Chinese-powered AI agents displaying deceptive behavior, unprompted replication and barrier circumvention in testing. And an arXiv paper reproduces the OpenAI–Hugging Face incident with publicly available models, noting that the compute budget largely determines which misaligned behaviors you can elicit.
"Voluntary" plus "no deadline" plus "we pick the auditor" is not a compliance regime, it's a vibe. The agents, notably, did not sign anything.
The Quiet One: Crypto.com's Agent Is Still in Stealth Mode
Worth flagging for contrast. Per Cryptonews.net, Crypto.com's AI agent is still in "stealth mode" nearly 8 months after its Super Bowl debut. Robinhood shipped agentic accounts in May, extended them to crypto in July, and now has 150,000 of them. One of these companies is treating agents as a product; the other is treating them as a teaser.
Also in the agent-token bucket: SOON jumped 31% on AI agent plans, per Altcoin Buzz. A headline and a chart, mostly.
Scam Desk: The SEC vs. "AI Signals"
The SEC charged four entities over two alleged schemes that used WhatsApp group chats and fake crypto platforms to take at least $15 million from retail investors, per Unchained. The Cryptoaiml complaint alleges chats posed as real finance professionals and shared supposedly AI-generated trading "signals" boasting 98% accuracy. Balances showed steady gains; the SEC says no trades occurred. Four main wallets collected nearly $12 million from more than 300 victims.
"AI-generated signals with 98% accuracy" is the same sales pitch as "my cousin works at Goldman," just with a GPU bill. If an agent's track record lives only on the platform that sold it to you, it isn't a track record.
Market Side-Eye
- Altcoin exchange deposits hit a seven-day total of 78,000 transactions on Sept. 28, up about 160% from Sept. 14, with depositing addresses nearly tripling to 51,600 — the highest since October 2025, per CryptoQuant, which notes holders usually move coins to exchanges to sell.
- Bitcoin ETFs extended their net inflow streak to nine trading days, pulling in roughly $3.1 billion, while Ether and Zcash funds flipped to outflows, per Cointelegraph. BTC traded near $83,567, and the Fear & Greed Index slipped to 71.
- Bitget hackers moved about $3.9 million in ZEC (2,746 ZEC) into Zcash's Ironwood shielded pool, roughly 15% of the ZEC taken in the $387.5 million breach, per CoinDesk.
- Standard Chartered initiated coverage of Ethena, forecasting ENA at $2 by end-2028 (roughly 669% upside from $0.26) and USDe supply growing to about $40 billion, per The Block and Unchained.
- Senator Blumenthal called Tether's USDT a "superhighway" for Iranian sanctions evasion, citing a subcommittee analysis of 846 sanctioned wallets, 84% of which transacted exclusively or nearly exclusively in USDT, per Decrypt.
Closing Take
Robinhood just made the AI-agent pitch concrete: real accounts, real leverage, real money, shipped to retail before any regulator has decided what an "agent" even is. The White House deal, meanwhile, is a handshake with a mirror. The gap between how fast these things get productized and how slowly they get audited is the entire story — and it is not closing.
Not financial advice. Side-Eye Signals reports what sources say, not what you should buy.