What if Citi used HarmCheck?
An illustrative analysis based on Lindsey v. Citigroup Global Markets, Inc. (S.D.N.Y., Nov. 20, 2023). Every example below is drawn directly from the public federal complaint. We map the alleged conduct to the Active Listening classifiers that would have flagged it — at the moment a message is drafted, before it ever leaves the firm.
A “locker room” environment, hiding in plain text.
On November 20, 2023, a 15-year Citi Managing Director filed a federal complaint in the Southern District of New York alleging years of harassment, sexual assault, and retaliation inside the bank’s Global Equities division (Lindsey v. Citigroup Global Markets, Inc., No. 1:23-cv-10166).
The conduct described in the complaint is not subtle. According to the public filing, a former senior executive sent the plaintiff hundreds of abusive messages — explicit threats of violence against her and her children, threats to cut her compensation, threats to sabotage her reputation, and degrading sexualized commentary. The wider environment alleged in the complaint — ranking female employees by appearance, mocking the firm’s harassment trainings, excluding women from client opportunities — extended well beyond a single bad actor.
Every one of those categories — violent threats, retaliation, harassment, hostile environment, quid-pro-quo coercion — is a named Title VII, EEOC, and FINRA risk surface. And every one is something Active Listening reads for, by design.
- Explicit threats of violence — e.g., “And I am going to set you on fire” (¶6)
- Threats targeting family — e.g., “Your kids’ life will be ruined from here on in” (¶6)
- Retaliation against compensation and career — e.g., “Taking you down in comp. Hard.” (¶6)
- Reputation and reprisal threats — e.g., “I plan to f*** your life… drop something on social media and tag the community.” (¶6)
- Sexualized harassment — repeated alleged comments on female employees’ bodies and appearance (¶2, ¶6)
- Pattern, not incident — alleged misconduct across 15 years, multiple supervisors, one division
Every category here is a classifier HarmCheck was built for.
HarmCheck Active Listening sits inside Outlook — reading every draft before it leaves an employee’s keyboard. It runs ~50 proprietary classifiers tuned to regulatory, employment, and conduct risk. The categories alleged in the Lindsey complaint are not edge cases; they are exactly the categories the system was purpose-built to detect.
When language patterns like the ones cited in the complaint hit the system, two things happen: the employee is shown a flag and an explanation of the violation, and is given an opportunity to change the message. If the message is sent without removing that flag, it is held in quarantine and is only released if a compliance admin allows it.
- Violent Threats — language indicating intent to cause physical harm to a person or their family.
- Threats Against Family / Coercion — messages targeting an individual’s children, relatives, or dependents to intimidate or compel behavior.
- Retaliation & Coercion — explicit or implicit threats to a person’s compensation, role, reputation, or professional standing in response to a complaint, rejection, or refusal; also language conditioning a workplace benefit, opportunity, or status on sexual or personal compliance, especially in supervisor-to-direct-report communications.
- Sexual Harassment / Hostile Environment — sexualized commentary, unwelcome propositions, or degrading remarks based on sex, gender, or sexual orientation.
- Restricted-List violations / MNPI — communication of material non-public information, or contact across information barriers maintained on the firm’s restricted list.
What changes when these categories hit a draft, not a courtroom.
Drafts containing violent threats, retaliation language, or sexual harassment are stopped before they leave the firm. The audit record — drafter, recipient, channel, classifier, timestamp — is created in the same moment, so compliance has evidence of intent, escalation, and pattern from day one rather than discovering it years later in a federal complaint.
When patterns repeat across senders or escalate over time, Active Listening surfaces them as conduct risk before they become a 15-year scandal across a trading floor. Investigations that once required outside counsel and Rapid Deploy retrospectively can instead happen in real time, on the firm’s own infrastructure, with the audit trail already built.
That is the precise gap Active Listening closes. The channels themselves become the signal — so compliance gets visibility without depending on a victim being brave enough to walk into HR knowing the people who would retaliate are the same people they would be reporting.
- Threats and harassment intercepted at the moment of draft — not discovered in litigation discovery
- Pattern of misconduct surfaced as a single conduct-risk view across senders, channels, and time
- Audit-ready record for every flagged draft — original message, triggering sentence, category, classifier version, reviewer action
- Hundreds of millions in potential settlement, regulatory, and reputational exposure addressed before it ever lands in a federal docket
“Women… reasonably understood that if they complained, the Bank would retaliate against them, because senior managers were involved in making and condoning these harassing comments.”
More case studies
Fair housing case: 246,000 documents reviewed in six days.
In a fair housing suit against a city council accused of blocking a housing development, discovery produced three years of communications. Rapid Deploy ran the corpus at 20,000+ pages per hour — surfacing coded NIMBY language and pretext patterns with sentence-level citations.
Bulge-bracket bank stops an MNPI leak before send.
An equity research analyst began drafting a market commentary referencing a name on the firm’s restricted list. Restricted List intercepted the message, blocked the send, and routed it for compliance review — before any external party saw it.
See HarmCheck in your environment.
Every case study starts with a conversation. Tell us what you’re trying to protect, and we’ll show you exactly how HarmCheck would have caught it.
Request a Demo