Cubist Systematic Strategies, the quantitative-trading arm of Point72, alleges in a lawsuit that a former research analyst issued 4,117 queries across 2,416 sessions with Claude Code and OpenAI Codex between February 28 and May 31, 2026, to distill colleagues’ research and source code into portable files. Bloomberg first reported the suit on 10 November.
The verified complaint, dated 9 November and filed in Connecticut Superior Court at Stamford, says both coding assistants were “Company-provided”, and that Cubist recovered the original AI usage logs after the analyst deleted the AI history. By our arithmetic, the window is 93 days, or about 44 queries a day. These are allegations and have not been tested in court.
Business Pill · Least Privilege
An evergreen Business Engineer explainer on giving an AI agent only the access its task needs. It teaches the general idea only and does not discuss Cubist or this case.
The key insight: As we read it, the complaint describes coding agents doing what they are built to do, reading whole repositories on request, and the firm’s own prompt logs as the record of it: 4,117 queries in 93 days, about 44 a day by our arithmetic, all allegations not yet tested in court.
According to the complaint, the analyst worked on Cubist’s KEPL team, a group of about 14 people that trades medium-frequency statistical arbitrage, from December 2019 and was later promoted to Research Analyst Lead. Cubist says it terminated the analyst for cause on September 18, 2026, after an investigation.
The complaint alleges the analyst used the two assistants to “access and review the proprietary research and source code” colleagues had developed, “distill that research and code into portable, easily readable markdown files”, test whether those files were enough to rebuild the work outside Cubist’s platform, and take them off company systems.
It says “nearly half” of the 4,117 queries were made outside normal working hours or at the weekend. It also alleges the analyst photographed proprietary source code on Sunday, May 3, 2026, while interviewing with competitor firms, and that documents the analyst later handed over held markdown files distilling more than 20 of Cubist’s proprietary forecast models.
AI coding-assistant activity Cubist alleges in its verified complaint of 9 November 2026: 4,117 queries across 2,416 sessions with Claude Code and OpenAI Codex between 28 February and 31 May 2026, including more than 600 exhaustive-read prompts. About 44 queries a day is our arithmetic. These are allegations and have not been tested in court.
How the AI Tools Figure
The complaint draws a line between chatbots and coding agents: “Unlike typical chatbot-style agents such as ChatGPT, these AI coding assistants have direct read and write access to code repositories”. It says the analyst directed Claude Code to read and extract confidential information from 14 different codebases, “nearly every repository of the KEPL team”.
In total, the complaint says, the analyst issued “over 600 prompts that used exhaustive-read language”, asking the tool to “extract all” and give “maximum detail”. It quotes a March 21, 2026 prompt that asked for “every formula, every regression specification, every parameter value, every key finding” from a colleague’s research notebooks, framed as being “for writing an academic paper”.
The same logging is how the case was built. The complaint lists, among Cubist’s safeguards, “Monitoring employee AI usage, including logging an employee’s AI prompts and the output from those prompts”. Earlier this month we reported on Apple, Goldman Limit AI as OpenRouter Token Prices Fall 57%, another case of finance firms setting rules for coding tools.
The complaint’s 4,117 Claude Code and Codex queries, split at its “over 600” exhaustive-read prompts. Shares are our arithmetic at 600; the allegations have not been tested in court.
Our Numbers
February 28 to May 31, 2026 is 93 days counting both ends, so 4,117 queries is about 44 a day, by our arithmetic. Across 2,416 sessions, that is about 1.7 queries a session, by the same arithmetic.
The more than 600 exhaustive-read prompts are more than one in seven of the 4,117 queries, about 14.6% at 600, by our arithmetic. The complaint says a separate 1.5-year “study plan” the analyst asked Claude Code to draft matches the eighteen-month non-compete, which Cubist asks the court to enforce from the termination date.
The Structural Read
As we read the complaint, the case rests less on the tools than on access and logging: it says the analyst could open colleagues’ private repositories because of a review role, and that Cubist rebuilt the activity from AI usage logs it recovered after deletion.
As we read the relief Cubist asks for, the eighteen-month non-compete, a forensic inspection of personal devices and damages in an amount to be proven, the firm is treating distilled markdown files as copies of its code, not as notes.
Cubist Systematic Strategies verified complaint, 9 November 2026
“Unlike typical chatbot-style agents such as ChatGPT, these AI coding assistants have direct read and write access to code repositories”
Three Implications
FOR FIRMS GIVING STAFF CODING AGENTS The complaint lists logging an employee’s AI prompts and the output from those prompts among Cubist’s safeguards, and says that is how the activity was reconstructed.
FOR REPOSITORY ACCESS Cubist says the analyst’s access to colleagues’ repositories was provided for limited purposes, such as reviewing junior team members’ code.
FOR THE CASE Cubist asks for a temporary and permanent injunction, a court-approved forensic inspection and damages in an amount to be proven; the complaint’s return date is December 15, 2026.
The framework says: “The surface, not the model, governs autonomy.”
As we read it, the Cubist complaint is that idea seen from the security side: the harness, here a coding agent with read access to a repository, set how much one user could pull out. The framework predates the suit and does not discuss Cubist or Point72.
What Is Not Established
The complaint is one side’s account. The court’s docket shows the former analyst’s lawyers filed an objection to ex parte relief on 10 November. eFinancialCareers reported that the analyst’s lawyer told Bloomberg the accusations are without merit, and that in an initial filing the analyst said “every relevant device and account has been forensically imaged and is preserved” and that Cubist owes more than $30 million “in investments and compensation wrongfully withheld”. We have not read that filing, and we did not contact the parties.
The complaint does not name the rival firm, and it seeks compensatory damages “in an amount to be proven” rather than a figure; it says Cubist “cannot estimate with any certainty” its damages. Neither Anthropic nor OpenAI is a party, and the complaint does not allege any fault by either tool.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Point72’s Cubist alleges a former research analyst used company-provided Claude Code and Codex for 4,117 queries in 93 days, about 44 a day by our arithmetic, to distill and take its research and code. The firm wants an injunction, a forensic inspection of personal devices and damages in an amount to be proven; the allegations have not been tested in court.
A note on sourcing. We read the verified complaint Cubist Systematic Strategies filed in Connecticut Superior Court at Stamford, dated 9 November 2026, in full (its exhibits excepted), from the copy Bloomberg linked, the opening of Bloomberg Law’s report, the court’s public docket and eFinancialCareers’ report of the former analyst’s response. Everything said about the former analyst’s conduct is an allegation by Cubist that has not been tested in court; we do not name the individual. Day counts, rates and shares are our arithmetic. We did not contact Cubist, Point72 or the former analyst for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
