Grok Autopilot portfolio up 59%, ahead of Claude fund

Grok Autopilot’s autonomous stock portfolio returned 59% over its first nine months, compared with a 36% rise in the S&P 500 for the same period, according to public Autopilot data. About $17 million is currently mirrored to the Grok account on the platform.

The two experiments run by AI Finance Labs on the Autopilot mirror-trading platform are visible to the public as @grkportfolio and @theaiportfolios. The wider Autopilot lineup holds roughly $150 million in mirrored capital across strategies.

Over the most recent three months the Grok account added about 12.6%, while the SPY ETF rose about 9.75%. Public posts show Grok’s portfolio has concentrated positions in AI infrastructure, semiconductors and memory, with additional exposure to energy and defense stocks that helped during market shocks earlier in 2026. The account posts indicate it operates without human override.

Claude’s AI-managed fund started in April 2026 with $50,000 in seed capital and also executes trades without human intervention. Operators posted that after about two months SPY was up roughly 8.3% while Claude’s returns were near 2.6%.

Public trade logs show the Claude account rotated into ServiceNow and Zeta Global and trimmed Microsoft in recent weeks, reflecting a tilt toward enterprise software, fintech and power. The operators have described the agent’s process as using probability-weighted scenarios, kill conditions and forward catalysts to justify trades.

The public nature of the accounts creates a direct comparison of how different AI agents convert market data into trades. Grok’s heavier exposure to hardware and energy contrasts with Claude’s more hedged exposure to software and services.

Anthropic has received regulatory warnings about use of its name on retail products, and independent traders outside the Autopilot experiments have used Anthropic models to run automated betting and trading bots.

Raullen.eth, an AI developer and active user on X, wrote that “They lack real intelligence, so expecting them to trade and consistently beat humans in a reasonable timescale doesn’t make sense.”

Investors mirroring either account face fees and concentration risk. Upcoming earnings reports and sector rotations will provide further data on whether infrastructure and hardware bets or defensive software and services exposures perform better in live trading.

The two public Autopilot portfolios will continue to serve as live experiments for how different AI decision-making frameworks manage actual capital.