Curriculum·R410 Copy Trading, Vaults, and Agentic Execution·about 33 min
Agentic execution and AI trading
By the end of this lesson you can
- →Separate a claim about a method from evidence about a result
- →State why a model's description cannot be verified from outside, and what can
- →Compute the sample an agentic system needs before its expectancy is distinguishable from zero
- →Apply the permission controls that bound an autonomous system's authority
Senior · enrolled learners
This lesson opens with The SEC's AI washing orders against Delphia and Global Predictions, 18 March 2024.
- What happened
- On 18 March 2024 the SEC announced settled charges against two investment advisers, Delphia (USA) Inc. and Global Predictions Inc., for making false and misleading statements about their use of artificial intelligence. Global Predictions had described itself in 2023 as the first regulated AI financial advisor and advertised expert AI-driven forecasts. Delphia, a robo-adviser, claimed from 2019 that it used artificial intelligence and machine learning on its clients' own data, such as social media and spending, to make intelligent investment decisions and predict which companies and trends were about to make it big. Its investing algorithms in fact never used client data at all. According to the SEC's order, Delphia admitted to SEC examiners in July 2021 that it had not used any of its clients' data and had not created an algorithm to use client data, and it continued to make false statements about using client data through August 2023, including a 2022 post that its proprietary algorithm used the data invested by its members. The firms paid civil penalties of $225,000 and $175,000 respectively, $400,000 in total, without admitting or denying the findings. The law firm Debevoise & Plimpton described them as the SEC's first cases charging antifraud violations over AI disclosures.
- The decision point
- Delphia told SEC examiners in July 2021 that the client-data capability did not exist and kept advertising it in some statements for a further twenty-five months, which is possible because a claim about a method cannot be checked by a client. No client could inspect the model, and it took a regulator's examination to find the gap. What a client could have checked was the record the system produced, which per R405-04 requires a sample and per part three is the only evidence about an automated system available from outside.
- Recorded loss
- $400,000
What you will be able to answer
- →What can and cannot be verified?
- →How long did the false claim persist?
- →What sample does an agent need?
- →What bounds an autonomous system's authority?
Orientation and Year One are open: anyone can read them without an account. From Year Two onward the lessons are for enrolled learners, because progress through the later years only means anything if it is tracked against a record.
It is free. We do not sell the list and there is nothing to buy at the end of it.
Sources and review
- https://www.sec.gov/newsroom/press-releases/2024-36
- https://www.sec.gov/files/litigation/admin/2024/ia-6573.pdf
- https://www.wilmerhale.com/en/insights/client-alerts/20240327-sec-brings-two-more-ai-washing-enforcement-actions-against-investment-advisers-continuing-its-pursuit-of-misstatements-related-to-ai
- https://www.debevoise.com/insights/publications/2024/03/ai-enforcement-starts-with-washing-the-sec-charges
- https://www.bleepingcomputer.com/news/security/crypto-platform-3commas-admits-hackers-stole-api-keys/
- https://www.fca.org.uk/news/press-releases/fca-fines-cgml-27-million
- https://www.sec.gov/newsroom/press-releases/2013-222
- https://www.sec.gov/Archives/edgar/data/0001060749/000119312512332176/d391111dex991.htm
Confidence high·Volatility high·Reviewed 2026-10-22·Owner unassigned
Contested
Delphia did run investing algorithms built on third-party data; what the SEC's order found false was the claim that client data and AI or machine learning on that data fed them. The 'first' characterization is Debevoise & Plimpton's, not the SEC's press release, and other law firm commentary at the time described the pair as further AI-washing actions rather than the first. The Knight figure is the SEC's; Knight's own announcement at the time put its pre-tax loss at about $440M.
Both matters were settled without admitting or denying the findings recited in the orders, and the penalties were small relative to the sums in most cases in this course. They are used here because they are the clearest documented instance of the specific gap this lesson describes, which is that a claim about a method is unverifiable from outside. Nothing here is a claim about any other firm.
This lesson makes no assessment of whether machine learning methods can produce a trading edge. Per P6 that is an open question with serious work on both sides, and per R405-04 it is answerable only through out-of-sample records rather than through descriptions. The lesson is about what evidence is available to you.
F109-04 owns delegated signing authority and R409-02 owns the API key exposure on bot platforms. This lesson owns the evidentiary question and the permission bounds specific to autonomous execution. Keep the splits.