Mollick's judgment: After obtaining AGI, top-tier labs might first hide away to focus on financial arbitrage.

Title

Ethan Mollick: If You Already Have AGI, Why Tell Others?

Summary

Wharton School professor Ethan Mollick made an uncomfortable assertion on X: when a lab develops human-level AGI, the fastest and most stable way to make money is not to launch an API for everyone to use, but to first leverage this intelligence for trading in financial markets—taking advantage of competitors before they even realize what’s happening to secure excess returns.

Following this logic, true technological breakthroughs are more likely to be hidden rather than publicly announced.

This contradicts the premise of mainstream AI governance discussions. There is a common assumption that “we will know when the crucial moment arrives.” But according to Mollick, we may not know at all.

Analysis

• Who is Mollick: Associate Professor at Wharton, head of Generative AI Labs, author of the upcoming bestseller “Co-Intelligence” in 2024, his Substack column “One Useful Thing” has over 419,000 subscribers, and TIME named him one of the influential AI figures of 2024. His analysis of industry incentive mechanisms carries enough weight to be taken seriously.
• Similar concerns have been raised elsewhere:
  • The AI Policy Bulletin (2025-2026) warned about “the obscurity of AGI training”—to evade regulation or gain a head start, labs may underreport their disclosures, leaving no public oversight when issues arise.
  • JPMorgan’s (2025) research on AGI and finance also supports this premise: if there is a system that comprehensively surpasses human cognitive tasks, using it for quantitative trading and strategy discovery would be highly enticing.
  • Anthropic’s research on “malicious behavior of agent systems in corporate scenarios” also suggests: “hiding to earn a profit first” is not a delusion, but a rational response to incentive mechanisms.

Choosing Between Two Paths: Open API vs. Secret Trading

Conclusion: If the goal is to make the most certain money in the shortest time, the motivation to hide and arbitrage in financial markets is stronger.

Implications for Governance

• Assumptions may be wrong: Most governance frameworks default to the notion that “breakthroughs are observable,” but Mollick’s scenario indicates that “unobservable breakthroughs” align better with incentive logic.
• Regulatory invisibility: Even with AI disclosure requirements, as long as there are no mandatory independent audits and continuous monitoring, labs have ample space to create an “information buffer” between technical capability and compliance.
• What the market would look like: If someone is secretly using AGI for trading, price signals will first be internalized by a few, and then only manifest in the public market as abnormal volatility and noise.

Impact Assessment

• Importance: High
• Category: AI safety, market impact, AI research

Judgment: For ordinary traders, this news may already be too late; the true beneficiaries are the labs that grasp the system first, along with proprietary firms and hedge funds that have private data and strategic capabilities. For regulators and infrastructure builders, there is still an opportunity to layout tools for “mandatory auditing + behavioral monitoring” early on. Overall, the advantage clearly leans towards players who “possess intelligence and can execute in a closed manner.”