BlackRock Aladdin vs. Citadel Securities: What We Actually Know About AI in Institutional Trading

Corrected version — published August 25, 2026

Editor's note: This article replaces an earlier version that contained unverifiable statistics and a fictitious Citadel Securities product name ("Cedar"), along with regulatory citations ("Regulation AI-FID," an EU AI Act "Systemic Risk Tier for HFT") that could not be traced to any SEC, ESMA, or European Commission source. This version retains only claims that could be verified against primary or highly credible secondary sources, listed inline. Where evidence was insufficient for a direct comparison, the framing has been changed accordingly rather than filled in with invented figures.

Key Takeaways

• BlackRock has built a documented, named AI system — Aladdin Copilot — into its Aladdin platform, using a supervised agentic architecture (LangChain/LangGraph with GPT-4 function calling) to help users query the platform in natural language. This is confirmed by a BlackRock engineering presentation cited in ZenML's LLMOps case study database.
• BlackRock has also integrated climate and ESG risk analytics into Aladdin through partnerships with RepRisk and Clarity AI, layering AI-driven satellite, news, and disclosure data onto portfolio risk analysis.
• Aladdin's Annual Contract Value (ACV) — the standard metric BlackRock reports for its technology business — grew 30% year-over-year in Q1 2025 (14% organically), per BlackRock's own earnings call transcript. No BlackRock disclosure attributes a specific dollar figure to "AI bookings" alone, and no verified figure for Q1 2026 could be located at the time of writing.
• Citadel Securities has publicly confirmed it has used machine learning in its trading operations for roughly a decade, starting with TensorFlow and later transformer-based models, according to remarks by CEO Ken Griffin reported by Global Trading (July 2026). Griffin also described an internal agentic AI system that can reproduce and verify academic finance papers in two to three hours, a task he said previously took six to eight weeks.
• No evidence was found for a Citadel Securities product called "Cedar," for a deep reinforcement learning execution engine simulating "millions of counterfactual scenarios every 100 milliseconds," for a 19-basis-point market impact reduction, or for $4.5 billion in client crossing-cost savings. These claims do not appear in Citadel Securities' own newsroom, SEC filings, or in any Reuters, Bloomberg, or FT reporting located during fact-checking. They have been removed rather than replaced.
• No SEC rule called "Regulation AI-FID" or "Reg 210.30" exists. The closest real regulatory development is the SEC's 2023 proposed rule on "Conflicts of Interest Associated with the Use of Predictive Data Analytics by Broker-Dealers and Investment Advisers," which has not been adopted as of this writing, plus a separate and distinct enforcement track against "AI washing" (misleading claims about AI use), which has produced settled cases (e.g., Delphia USA and Global Predictions, March 2024).
• No "EU AI Act Systemic Risk Tier for HFT" was activated on March 1, 2026. Under the actual EU AI Act, "systemic risk" is a classification applied to general-purpose AI models above a compute threshold — it is not a trading-specific tier. Separately, ESMA published a non-binding supervisory briefing on algorithmic trading under MiFID II on February 26, 2026, which addresses firms' use of AI in trading algorithms but explicitly notes that AI in trading is not currently classified as "high-risk" under the AI Act.

What BlackRock Has Actually Disclosed

BlackRock's Aladdin platform is the firm's end-to-end risk management, portfolio management, and operations system, used internally and licensed to institutional clients across more than a dozen years of public disclosure. In 2025 and 2026, BlackRock layered generative and agentic AI onto this base:

• Aladdin Copilot, per a technical presentation by BlackRock's AI engineering team (Brennan Rosales and Pedro Vicente Valdez) documented by ZenML, is a supervised multi-agent system that lets users query Aladdin's hundreds of domain-specific APIs in natural language, with guardrails built for regulated financial-services use.
• Aladdin Climate, built in part on a 2021 technology partnership with Baringa Partners and later enhanced through a partnership with Clarity AI, models physical and transition climate risk (e.g., extreme weather, policy shifts) at the instrument and portfolio level. Coverage from FinTech Futures and Medium-published industry analysis (July 2025) describes the integration of RepRisk's AI-driven ESG data feeds, though these are trade-press accounts rather than BlackRock press releases, and are presented here as such.
• On financial performance, BlackRock's own quarterly earnings calls are the authoritative source. Recent transcripts show technology services and subscription revenue up 16% year-over-year in Q1 2025, with Aladdin's ACV up 30% year-over-year (14% organically) — driven by "sustained demand for our full range of Aladdin technology offerings," per CFO Martin Small, and partly by the Preqin acquisition closing March 3, 2025. BlackRock does not break out a separate "AI-attributed" revenue or bookings figure in these disclosures, so any such number should be treated as unverified until a specific filing or release states it explicitly.
• Trade press (STP Investment Services/FundFire, May 2026) describes BlackRock continuing to expand Aladdin's agentic capabilities into "exception management, compliance interpretation, and portfolio decision support" through 2026, citing BlackRock executives — a directionally credible but qualitative claim, not a quantified one.

What Citadel Securities Has Actually Disclosed

Citadel Securities is a market maker, not an asset manager, so its AI use sits in a different part of the investment lifecycle: pricing, liquidity provision, and trade execution rather than portfolio construction. Verified points:

• CEO Ken Griffin has stated publicly (Global Trading, July 2026) that Citadel and Citadel Securities have used machine learning for approximately a decade, moving from TensorFlow to transformer-based models to improve "pricing, trading and risk management." He also described building an agentic AI system that compresses academic-paper reproduction from six-to-eight weeks down to two-to-three hours, and said the firm has not cut staff as a result of AI productivity gains, redirecting people to new opportunities instead.
• Citadel Securities publishes a regular macro-strategy research series ("Global Market Intelligence," "Outpacing Change") through its own newsroom, which references AI adoption trends in the broader economy but does not disclose execution-specific AI performance metrics such as basis-point cost savings.
• Independent industry summaries (e.g., Appinventiv, February 2026) describe Citadel Securities' ML use in general terms — processing large volumes of tick data to react to price changes and refine market-making — consistent with, but less specific than, Griffin's own remarks. These are treated as directional confirmation, not as a source for hard figures.
• No named "Cedar" system, no disclosed reinforcement-learning execution engine, and no specific execution-quality statistics (basis points, dollar savings) were found in any Citadel Securities source, SEC filing, or reputable financial press report. If such a system exists, it has not been publicly documented as of this writing.

The Actual Comparison: Risk/Portfolio AI vs. Market-Making AI

Given the evidence available, the defensible comparison is not "whose AI performs better" but where in the investment lifecycle each firm applies AI:

This is a comparison of disclosure practices and business models as much as of technology. BlackRock, as a public company with a large SaaS-like technology business, reports Aladdin financial metrics quarterly and has published case-study-level technical detail about Aladdin Copilot. Citadel Securities is privately held and does not report execution-quality statistics publicly; what is known about its AI use comes mainly from its CEO's on-the-record comments rather than filings or product documentation.

Regulatory Landscape (Verified)

• The SEC has been active on AI in investment advice since at least 2023, when it proposed (but has not adopted) a rule on conflicts of interest from "predictive data analytics." Separately, the SEC has brought and settled "AI washing" enforcement cases — most notably against Delphia (USA) Inc. and Global Predictions Inc. in March 2024 — for overstating AI capabilities to investors. The SEC's Division of Examinations listed AI-related risk among its FY2026 examination priorities.
• In the EU, ESMA published a supervisory briefing on algorithmic trading under MiFID II on February 26, 2026, which extends existing algorithmic-trading governance and testing expectations (per RTS 6) to cover AI-influenced trading systems. Separately, an ESMA/EIOPA/EBA joint statement (2026) references the EU AI Act's "systemic risk" classification for general-purpose AI models — a compute-threshold-based classification, not a trading-specific tier, and not something "activated" on a specific date for HFT firms as such.
• Neither regulator has adopted a rule requiring a specific percentage-of-risk-budget threshold for human-in-the-loop review of AI recommendations, as far as could be verified. Any such figure should not be treated as established regulation absent a located primary source.

What Remains Genuinely Uncertain

• Whether BlackRock's AI features have driven a specific, isolated revenue or bookings number — as opposed to overall Aladdin ACV growth — is not established by any source located during fact-checking.
• Whether Citadel Securities operates a named, DRL-based execution system comparable in scope to what the original article described is unverified; it may exist without public disclosure, but no source confirms this either way.
• Broader claims common in this space — such as McKinsey or Gartner estimates of AI-driven value creation across the investment industry — were not independently re-verified for this rewrite and have been removed rather than carried over uncritically. Readers interested in those figures should consult the original McKinsey/Gartner publications directly.

Bottom Line

The verifiable story is narrower than the "autonomous alpha vs. autonomous execution" framing suggests. BlackRock has documented, technically detailed, and financially disclosed AI integration into Aladdin, concentrated on natural-language access to risk and portfolio data plus climate-risk modeling. Citadel Securities has a long-standing, CEO-confirmed history of applying machine learning to market making, plus a notable and specific claim about agentic AI accelerating internal research — but has not publicly disclosed a comparably detailed execution-AI product or performance statistics. Claims of a head-to-head "19 basis points vs. $1.2 billion" comparison, and the regulatory framework cited to justify urgency around them, do not hold up against primary sources and have been removed.

Sources consulted: BlackRock Q1 2025 earnings call transcript (MarketBeat); ZenML LLMOps Database case study on Aladdin Copilot; FinTech Futures (Clarity AI partnership); BusinessWire (Baringa Partners partnership); STP Investment Services/FundFire (May 2026); Global Trading (Citadel/Griffin remarks, July 2026); Citadel Securities newsroom (citadelsecurities.com); Appinventiv industry overview; SEC.gov press releases and JD Supra legal-alert aggregation on AI-washing enforcement and the 2023 predictive data analytics proposal; ESMA supervisory briefing on algorithmic trading (February 26, 2026); ESMA/EBA/EIOPA joint statement on frontier AI models (2026); Macfarlanes legal analysis of the ESMA briefing. Claims that could not be traced to any of the above, or to an equivalent primary source, have been removed rather than presented as fact.