Michael Saylor said ChatGPT helped him design the preferred stock financing model that enabled Strategy to raise about $15 billion for its Bitcoin-focused balance sheet.

ChatGPT helped shape Strategy’s financing plan

Strategy Executive Chairman Michael Saylor said he used ChatGPT to explore and develop a preferred stock financing model tied to the company’s Bitcoin strategy, according to an Aug. 6 Fortune report.

Saylor discussed the process during an interview with The Diary of a CEO. He said the AI chatbot helped him examine financing structures that Strategy later used to raise billions of dollars from investors.

“AI helped me create $15 billion,” Saylor said.

The figure reflects capital raised through Strategy’s preferred stock products and related financing rather than revenue generated directly by ChatGPT. The AI tool helped Saylor work through the structure, while investors, underwriters and company executives carried out the offerings.

Strategy has built a group of Bitcoin-backed preferred securities, including STRC, STRK, STRF and STRD. The products give investors different combinations of dividends, volatility and exposure to the company’s Bitcoin-heavy balance sheet.

Saylor urges workers to ‘harness the robots’

Saylor said AI is changing how individuals and companies create value. In his view, workers should focus on asking better questions and using machines to pursue ideas that would otherwise require more time or expertise.

“Don’t try to outwork the robots,” he said.

His comments frame AI as a tool that can expand human decision-making rather than merely replace repetitive work. Saylor argued that future entrepreneurs would gain an advantage by combining human judgment and creativity with the speed of AI systems.

The claim also provides a practical example of generative AI being used in U.S. corporate finance. However, ChatGPT-generated proposals still require legal, accounting, and regulatory review before a publicly traded company can use them to sell securities.

Strategy’s preferred shares carry different dividend obligations and risk profiles. Their performance remains closely connected to the company’s ability to manage its capital structure and its large Bitcoin position.

Strategy adjusts its Bitcoin treasury policy

Saylor’s comments come as Strategy has shown greater flexibility in managing its Bitcoin holdings.

An Aug. 3 Securities and Exchange Commission filing showed that the company sold 1,638 BTC for approximately $104.73 million between July 27 and Aug. 2. Strategy used $52.4 million to fund preferred stock dividends and $52.3 million to repurchase STRC shares.

The sale reduced Strategy’s holdings to 842,138 BTC as of Aug. 2. The company reported an aggregate purchase cost of $63.51 billion and an average acquisition price of $75,419 per Bitcoin.

On Aug. 5, Lookonchain identified another transfer of 1,030 BTC, worth roughly $66.14 million, from wallets it associated with Strategy. Strategy had not confirmed that the transfer represented another sale, and no later SEC filing had reduced its reported holdings when this article was prepared.

Strategy expands beyond Bitcoin financing

Strategy is also adding employee benefits as it develops its broader corporate operations. On Aug. 5, the company joined the Invest America Business Pledge and committed to contributing $250 annually to Trump Accounts for eligible children of its U.S. employees.

Eligible children born on or after Jan. 1, 2025, will also receive a one-time $1,000 company contribution matching the U.S. government’s initial deposit.

The next test for Strategy will be whether its preferred stock model can continue attracting investors while supporting dividend payments and protecting its Bitcoin holdings. Saylor’s account of using ChatGPT shows how AI influenced the model’s design, but its long-term performance will depend on capital-market demand, Bitcoin prices and Strategy’s execution.

Spot Bitcoin ETFs have seen a notable acceleration in demand over the past week, according to Bloomberg ETF analyst Eric Balchunas. Several major funds reported inflows on every trading day since a Coldcard wallet vulnerability exploit became public—an overlap that has sparked renewed discussion about whether some investors are reconsidering self-custody in favor of regulated products.

Balchunas’ tally attributes roughly $620 million in cumulative inflows to BlackRock’s iShares Bitcoin Trust (IBIT), Fidelity’s Wise Origin Bitcoin Fund (FBTC), Bitwise’s Bitcoin ETF (BITB), ARK 21Shares Bitcoin ETF (ARKB), and Defiance Daily Target 2X Long MSTR ETF (MSBT). His observations echo earlier streak reporting from Cointelegraph, which covered an ETF inflow run reaching similar magnitudes.

Key takeaways

• Bloomberg’s Eric Balchunas says multiple spot Bitcoin ETFs recorded daily inflows for the entire stretch since the Coldcard exploit.
• Balchunas estimates the combined inflows at roughly $620 million across named funds.
• TRM Labs linked the Coldcard incident to theft of more than $116 million in Bitcoin from over 5,200 wallet addresses.
• The timing has intensified debate over the relative operational risks of self-custody versus ETF custody through institutional providers.
• Crypto security concerns are evolving alongside more sophisticated cyberattacks, including AI-assisted exploits highlighted by industry reporting.

ETF inflow streak lines up with Coldcard exploit fallout

The current inflow momentum centers on a simple pattern: funds that track spot Bitcoin exposure have continued bringing in net new capital day after day following the weekend Coldcard exploit. Balchunas’ post on X points to inflows at IBIT, FBTC, BITB, ARKB and MSBT every trading day since the incident, totaling about $620 million.

While the overlap is striking, Balchunas was careful to avoid claiming causation. He said on X that the connection is unknown—adding that, “long-term I can’t imagine there aren’t some who migrate over.” That framing matters: investors may already be rotating toward ETFs for accessibility and compliance reasons, but the Coldcard event appears to have sharpened attention on how custody failures can materialize even when users follow “best practice” assumptions.

For readers, the key question is whether this is a temporary spike tied to headlines—or evidence of a more durable shift toward ETF custody. The only way to judge that will be to watch whether inflows persist if attention on the exploit fades, or whether the streak breaks.

What happened in the Coldcard incident, and why it resonated

The Coldcard exploit involved a vulnerability affecting certain wallet devices, leading to significant losses. According to blockchain intelligence firm TRM Labs, the incident drained more than $116 million worth of Bitcoin from over 5,200 wallet addresses.

That scale is part of why the debate has reignited across the industry. Self-custody has traditionally been framed as a way to reduce reliance on intermediaries. But security research and real-world incidents have repeatedly shown that self-custody is not a single risk level—it’s a system of risks spanning device firmware, user setup, operational processes, and the broader ecosystem that supports hardware wallet usage.

The Coldcard episode thus serves as a reminder that hardware wallet users can still be exposed when flaws exist below the user interface—particularly when vulnerabilities emerge that can be exploited without requiring the user to willfully do something unsafe.

Self-custody vs. exchange custody: CZ revives the “statistical safety” argument

The ETF timing has also fed into broader arguments about whether self-custody is truly “safer” on a population basis. Binance co-founder Changpeng “CZ” Zhao commented on the ongoing discussion, suggesting that storing crypto on centralized exchanges may be “statistically safer” than self-custody, pointing to data from analyst Willy Woo. CZ’s reasoning is based on the visibility of losses: exchange-related incidents are easier for observers to detect and document, while self-custody failures (including hacks and lost funds) may go unreported or be harder to quantify.

In a post on X, CZ noted that “Hack data is easier to collect on the CEX side, usually major news. It is harder on the self-custody side, where hacks, lost coins, etc are often not reported.”

The practical implication isn’t that all custody models are equally reliable. Instead, it highlights an asymmetry in measurement: even if self-custody failures occur frequently, the public record can undercount them relative to highly visible exchange events. For investors deciding how to allocate Bitcoin exposure, this creates a problem of incomplete information—one reason ETFs continue to attract interest as a middle path between direct custody and exchange-managed holding.

Cyber risk is shifting, and custody debates are following

The Coldcard exploit arrives as concerns about rapidly escalating cyber threats become harder to ignore. Industry reporting cited by Cointelegraph points to increasing AI-assisted attack capabilities, where adversaries can identify and exploit vulnerabilities faster than defenders can patch them.

For example, Cointelegraph reported that Bitcoin swap service Boltz suspended its non-custodial bridge, citing a steady rise in AI-assisted exploits that were enabling attackers to move faster than the team could remediate issues. While that incident is not the same as the Coldcard hardware vulnerability, it reinforces a larger theme: attackers are increasingly benefiting from automation and speed—meaning the security burden on individuals and small teams can become disproportionately heavy.

That matters for custody decisions because self-custody security is often treated as a “set and forget” activity. In reality, device maintenance, software/firmware updates, environment hygiene, and broader operational discipline all require ongoing attention. If the threat landscape is accelerating, the gap between what users can comfortably manage and what attackers can probe may widen.

How investors should think about the ETF streak from here

Even if the Coldcard timing played a role in investor behavior, it may not be the only driver of ETF flows. ETFs already offer regulated access, standardized custody arrangements with institutional-grade processes, and simplified onboarding compared with direct device ownership and operational management.

The next test is persistence. Readers should watch whether daily inflows continue beyond the immediate news cycle and whether new inflow streaks emerge alongside future security incidents. If inflows remain strong while headlines fade, it would suggest that some capital is moving for structural reasons. If inflows taper quickly, the streak may reflect near-term sentiment shifts rather than a lasting change in custody preferences.

MetaMask has launched Agent Wallet, a self-custodial product that lets AI agents execute on-chain transactions while operating within limits set by users.

MetaMask Agent Wallet gives AI agents controlled access

MetaMask announced the wider launch of Agent Wallet on Thursday after introducing the product to roughly 200 early-access users in June.

The wallet targets traders and developers who use AI agents to monitor markets, identify opportunities and execute on-chain strategies. Supported activities include token swaps, perpetual futures trading, prediction market positions and liquidity provisioning.

Agent Wallet works with Claude Code, Codex, Cursor, OpenClaw, Hermes and OpenCode. Users can deploy agents across Hyperliquid and supported Ethereum Virtual Machine networks, according to MetaMask.

Unlike systems that give an AI model unrestricted wallet access, Agent Wallet lets users establish rules before funding the account. These include daily spending limits, approved protocols and broader risk preferences.

“This security-first model is why Agent Wallet is built around rules, not blind delegation,” MetaMask said. “The agent can act, but it acts inside the user’s boundaries.”

Users remain in control of their private keys and can export their secret recovery phrase, preserving the wallet’s self-custodial structure.

Guard Mode and Beast Mode offer different controls

Agent Wallet includes two operating settings based on how much autonomy a user wants to give an AI system.

Guard Mode is the default option. Transactions that exceed spending limits, interact with unapproved protocols, or violate another wallet policy pause for human approval through two-factor authentication. Users can review and approve the request through MetaMask Mobile or an email link.

Beast Mode reduces the number of approval interruptions for experienced traders and developers. Security checks remain active, while transactions identified as potentially malicious still require human authorization.

The wallet also uses gas abstraction. Agents can transfer or swap tokens without holding the network’s native asset to cover transaction fees. MetaMask instead settles the fee using the asset being moved.

“Agents can transfer and swap without holding a chain’s native token for gas; MetaMask settles the network fee in the token being moved.”

Transaction protection covers eligible losses up to $10K

Supported EVM transactions pass through simulation, threat scanning, and MetaMask’s Smart Transactions protection against maximal extractable value risks before reaching the blockchain.

Transactions that clear these checks may qualify for up to $10,000 in Transaction Protection coverage per month if they still result in a loss. The protection is subject to eligibility requirements and does not mean every trading loss, failed strategy, or unsupported transaction will receive reimbursement.

The distinction matters because autonomous trading remains exposed to market volatility, smart-contract failures, and incorrect instructions. Wallet-level limits can restrict what an agent is authorized to do, but they cannot eliminate every financial or technical risk.

MetaMask said the system is designed to prevent AI agents from overriding user policies, reducing reliance on the agent interpreting each instruction correctly.

MetaMask joins the race for AI-powered finance

The launch places MetaMask in competition with other crypto companies developing payment and trading infrastructure for autonomous software.

U.S.-based Coinbase introduced Agentic Wallets in February, giving AI systems programmable tools to spend, earn, and trade with built-in guardrails. MoonPay has also expanded MoonAgents, including a Telegram-based interface for managing crypto through natural-language commands.

Agent Wallet also extends MetaMask’s move beyond a standard browser wallet. In June, the company launched Money Account, a self-custodial account built around mUSD that combines variable DeFi yield, trading and card spending through one balance.

MetaMask has not introduced a native token alongside Agent Wallet. Adoption will instead depend on whether traders and developers are willing to give autonomous systems limited authority over real on-chain capital.

What have German officials said about the incident?