We Read Every SEC Enforcement Document Since 2010. Here Is What They Actually Charge.

Lately I have been getting a lot of requests to demonstrate how a local AI program actually works. So I built the demonstration: a local AI lab, the kind I walked through in What It Actually Looks Like to Run a Local LLM at Your Firm, loaded with a dataset I can share freely because it is public: the SEC’s entire enforcement archive, downloaded in full to local storage. The lab read all of it. This article is both the demonstration and the payoff, because what came back rewrites the standard story about SEC cybersecurity enforcement.

Most writing about this enforcement reacts to one order at a time, in the week it lands. That produces a distorted picture. The cases that get coverage are the ones with the biggest numbers, and the biggest numbers are almost never the cases that resemble your firm.

So we did it differently. The archive in the lab holds every administrative proceeding, litigation release, Commission opinion, ALJ initial decision, and ALJ order the agency has posted from January 2010 through today: 37,299 documents. The lab read the entire corpus for cybersecurity and IT-related conduct, pulling the charged provisions, the findings, and the penalties out of the order text itself; the independent verification pass then re-checked every claim in this article against the source documents.

To be precise about the method, because precision about AI claims matters in ways this article will make clear: we did not train a model, and we did not ask one what it remembered. Every claim below was read out of the source document, and every claim carries the SEC release number so you can pull the order and check us.

Two hundred nine documents matched, consolidating to 191 distinct enforcement matters: 102 in the off-channel messaging program, and 89 others that range from registered-firm safeguards cases to prosecutions of the hackers themselves. Set those prosecutions, the public-company disclosure cases, and the procedural orders aside, and roughly three dozen registered-firm cases carry the lessons that follow. And the single most useful thing the whole record teaches is this: there is no cybersecurity rule in these orders, and there is no case charging a registered firm for the fact of being breached. Not one, in sixteen years. The Commission charges four things, all of them documentary: safeguards policies that were not reasonably designed under Rule 30(a) of Regulation S-P, identity theft programs that did not function under Rule 201 of Regulation S-ID, failures to preserve electronic communications under Exchange Act Rule 17a-4 and Advisers Act Rule 204-2, and books and records that stopped being current after an incident. A breach is neither necessary nor sufficient. Three firms paid a combined $2.55 million in one day in 2022 with no intrusion alleged at all.

That finding is not trivia. It tells you exactly how to prepare, and it is the opposite of how most firms prepare.

Start with the examination, not the threat feed

Here is the philosophy this firm operates on, and the enforcement record is the strongest evidence for it we have ever assembled: cybersecurity compliance is an examination problem before it is a technology problem. The firms in these orders did not lose because an attacker was sophisticated. They lost because when the Commission asked to see the program their documents described, the program was not there. Every order in this corpus is, at bottom, a documented examination failure. Which means the record is not a list of scary outcomes. It is the answer key.

If you know precisely what an examiner will ask for, in what order, and what each document has to prove, then every case below stops being a cautionary tale and becomes a question you can answer in advance. That is the approach we walked through in What an SEC Examiner Actually Asks For: The Cybersecurity Document Request List, and it is the reason the Mock SEC Cyber Exam exists: thirty document requests drawn from real examination request lists, thirteen of them in the SEC’s own wording, that take about seven minutes to answer honestly. The output is not a score. It is your open items, mapped to the rule behind each request, in the order an examiner would find them.

Read the cases that follow with that lens. Each one is a request-list item the firm could not satisfy.

Where it started: a curb in Alpharetta

The earliest Safeguards Rule enforcement in the window did not involve a hacker. In September 2006, J.P. Turner customer records sat unsecured on a residential curb in Alpharetta, Georgia for roughly two weeks; a former branch manager had hired a company to pick up and destroy old files, the company never came, and he never followed up. An SEC examiner learned about it from a local news report, and the examination that followed moved from the curbside records to the firm’s written procedures. J.P. Turner litigated the resulting Rule 30(a) charge through a three-day hearing and lost: the administrative law judge found its policies and procedures were not reasonably designed to protect customer records, and the decision became the Commission’s final order in June 2010, with a cease-and-desist and a $65,000 penalty (Initial Decision Release 395, May 19, 2010; finality notice, Release 34-62313, June 17, 2010).

By April 2011 the Commission had reached individuals. GunnAllen Financial’s chief compliance officer was charged personally under the Safeguard Rule: after three laptops and a registered representative’s password credentials were stolen, he did not direct the firm to revise or supplement its procedures, and so aided and abetted the firm’s violation (Release 34-64220, April 7, 2011, $15,000). The firm’s president and its national sales manager were charged the same day under Regulation S-P’s privacy provisions for transferring customer records out of the collapsing firm without notice or opt-out (34-64221, 34-64222, $20,000 each).

The case that fixed the modern template for advisers came in 2015, and it was not brought against a wirehouse. R.T. Jones Capital Equities Management was a St. Louis adviser with roughly 8,400 client accounts and about $480 million in regulatory assets under management, without custody of client assets. To verify enrollment eligibility for its retirement plan product, it stored the personal information its plan sponsor partners sent over, covering more than 100,000 individuals, unencrypted on a third party-hosted web server. In July 2013 an intruder, traced to IP addresses in mainland China, gained full access and copy rights to the server; the log files were destroyed, so the forensic firms could never determine whether the data was actually taken.

The Commission’s charge was not the breach. It was that R.T. Jones had failed to adopt any written policies and procedures reasonably designed to safeguard the information: no periodic risk assessments, no firewall on the server holding client data, no encryption, in violation of Rule 30(a) of Regulation S-P (IA-4204, September 22, 2015, $75,000). The order also records what the firm did right: it promptly retained forensic firms, notified every affected individual, and offered identity monitoring, and no client is known to have suffered financial harm.

The lesson. Every element of the modern pattern is present before 2016: the violation is the missing program, the incident is merely the occasion, the response is priced into the outcome, and a CCO can be charged personally. The examiner’s first request is always the same one J.P. Turner failed: show me the written program. If yours needs building rather than defending, that is exactly what the Cybersecurity Policies & Procedures Manual in the BrainTrust library is for, and the P&P Manual Builder will draft one that is stack-specific, built from your answers.

The word that decides these cases is “designed”

Read enough of these orders in one sitting and a single phrase starts to repeat: not reasonably designed. It almost never means the firm lacked a policy. It means the policy said the right words and the firm could not show the words were operating.

Cetera is the cleanest example. Between November 2017 and June 2020, email accounts of more than 60 Cetera personnel were taken over, exposing personal information of over 4,388 customers. The Commission’s order notes that none of the compromised accounts had multi-factor authentication turned on, “even though Cetera Entities’ own policies required MFA ‘wherever possible,’ beginning in 2018” (Exchange Act Release 34-92800, August 30, 2021, $300,000). The policy was not the defense. The policy was the evidence.

M Holdings Securities is the recent restatement of the same principle, and the one small firms should study. M Holdings runs a network of 120 branch “member firms.” It had no network-wide written information security policy at all until September 2020, then adopted one requiring member firms to implement controls in 17 categories, including MFA, incident response, and security awareness training. The SEC charged the firm anyway, in November 2025, because as the order finds, M Holdings was aware that a significant number of member firms continued to lack the required policies and controls, and did not revise its own procedures to address that. Email accounts of 17 registered representatives and employees at 13 member firms were taken over; credential-harvesting email went out to roughly 8,500 people; four of the member firms were compromised twice (Release 34-104255, IA-6928, $325,000, charging both Reg S-P Rule 30(a) and Reg S-ID Rule 201). We took that order apart in detail in Practical Takeaways from the SEC’s M Holdings Cybersecurity Case.

Voya Financial Advisors shows what “not operating” looks like in real time. Over six days in April 2016, intruders called Voya’s own technical support line impersonating independent contractor representatives and asked for password resets to the customer-data portal. Two of the three reset requests came from phone numbers Voya had previously identified as associated with prior fraudulent activity. Support staff reset the passwords anyway and read temporary credentials over the phone, and on two of the three occasions supplied the username as well. Three hours after the first fraudulent reset request, the targeted representative reported the unrequested password-change email, and the intrusion continued anyway (Release 34-84288, IA-5048, September 26, 2018, $1,000,000).

The lesson. The examiner will not stop at whether MFA is required. The request will be for evidence it is enforced: enrollment reports, exception lists, the acknowledgements employees signed. A policy binder answers the first question; only records answer the second. The library carries an Access Control and MFA Policy for exactly this gap, and if you want to see how the account-takeover pattern plays out at a firm like yours, the KerberRose breach post mortem is the same fact pattern with 27,000 people on the other end.

No breach required

On July 27, 2022, the Commission settled with J.P. Morgan Securities ($1.2 million, IA-6073), UBS Financial Services ($925,000, IA-6074), and TradeStation Securities ($425,000, Release 34-95369) on the same theory: their identity theft prevention programs under Regulation S-ID did not comply with Rule 201. No breach is alleged in the orders. The programs existed on paper, but they did not function as programs. At J.P. Morgan and TradeStation, the red flags were copied from the illustrative examples in Appendix A of the regulation rather than derived from each firm’s actual covered accounts and incident history. UBS’s program, per its order, did not identify or incorporate any relevant red flags at all. None of the three updated their programs as risks changed.

That is worth sitting with. A boilerplate program is itself the violation. Robinhood drew the same Reg S-ID finding in January 2025 inside a broader settlement (Release 34-102170, $45 million in combined penalties across the two broker-dealer entities), alongside a Safeguards Rule finding that its procedures did not address a risk it had identified: unapproved remote access software on employees’ computers, which an unauthorized third party then used to reach customer records.

The lesson. This is the purest examination case in the record: three firms paid seven figures collectively because the document itself failed inspection, with no incident anywhere. If your Red Flags program was adopted years ago and reads like Appendix A, you are holding the same document they were. We wrote about why in Your Identity Theft Red Flags Program Is Probably a Dead Document, and the library’s Regulation S-ID Red Flags Policy was rebuilt this June around current-era red flags, including AI voice cloning, precisely because the old illustrative examples no longer describe how identity theft happens.

Vendors, disposal, and the fraud that walks in the front door

The largest pure Safeguards penalty in the record involves no intrusion. Morgan Stanley Smith Barney hired a moving and storage company with no data destruction experience to decommission two data centers. The vendor was supposed to work with an e-waste specialist to wipe the equipment; at some point it stopped, and sold roughly 4,900 devices, some containing unwiped customer data, to a downstream buyer. Separately, 42 decommissioned branch devices could not be located, and the encryption software on those devices had not been activated until 2018 (Release 34-95832, IA-6138, September 20, 2022, $35 million, charging Rule 30(a) and the disposal rule, Rule 30(b)).

Equiniti Trust, the transfer agent formerly known as American Stock Transfer, shows the same supervision failure from the impersonation side. A threat actor joined an existing email thread using a domain one letter off from a real client’s, imitated the client contact’s writing patterns, and directed the issuance of 5.3 million shares and the liquidation of 3.3 million of them, with $4.78 million wired to Hong Kong. The firm had circulated a warning email months earlier instructing employees to always call back on wire requests. The order’s finding is that the firm “did not confirm that the January 2022 warning email was read by its recipients, provide training to its employees on this topic, or otherwise ensure that call-backs were performed” (Release 34-100780, August 20, 2024, $850,000, charging Exchange Act Section 17A(d) and Rule 17Ad-12, the transfer agent safeguarding provisions, rather than Reg S-P). A second incident followed through a platform default that automatically linked accounts sharing a Social Security number even when names and addresses did not match. Net loss across both: about $4.08 million.

Under the amended Regulation S-P adopted in May 2024 (Release 34-100155), which took effect for smaller firms on June 3, 2026, service-provider oversight is no longer an inference from cases like these. It is written into 17 CFR 248.30 directly.

The lesson. An examiner working the amended rule will ask three things this section answers: show me your service-provider oversight procedures, show me your disposal procedures under 248.30(b), and show me the verification procedure that stands between an emailed instruction and a wire. The library carries a purpose-built document for each: the Vendor Due Diligence Questionnaire and the Vendor Contract Clause Rider with the 72-hour notification language, the Data Disposal Policy, and the Wire Transfer & Disbursement Verification Procedure. The thinking behind the first is in How a Spy Would Conduct Vendor Due Diligence, and behind the last in The Wire Transfer Call-Back.

The two-billion-dollar recordkeeping program

The largest IT-related enforcement effort in the record we reviewed was never about security. It was about where the messages lived. Beginning in late 2021, the Commission charged firm after firm whose personnel conducted business over personal text messages and messaging apps that were never captured by the firm’s archiving systems, in violation of Exchange Act Section 17(a) and Rule 17a-4 for broker-dealers, and Advisers Act Rule 204-2 for advisers.

Our corpus contains 103 off-channel documents naming 101 distinct respondent groups; the SEC’s fiscal year 2024 enforcement results release puts the initiative at more than 100 firms and $2 billion in penalties, a figure Commissioner Peirce cites in her dissent in Release 34-102860. The headline tier ran to $125 million per firm. But the tier that matters for readers of this site is the September 17, 2024 municipal advisor sweep, where total penalties ran from $40,000 to $324,000. This program reached firms of every size, and it reached RIAs; we covered that arm of it in Off-Channel Communications: The Sweep That Reached RIAs.

Two findings from the tail end of the program are worth more than the totals. First, cooperation was priced openly. PJT Partners self-reported its off-channel failures and settled for $600,000 on the same conduct theory that cost non-reporting firms eight and nine figures (Release 34-102167, January 13, 2025); the order credits the self-report by name. And outside the off-channel program, when ransomware-struck ICBC cooperated and remediated, the Commission imposed no civil penalty at all, as covered below. Second, the undertakings survived the change in enforcement climate. In April 2025, respondents under sixteen settled orders moved to soften their terms: replace the two-year independent compliance consultant with a one-time internal audit, drop the two-year obligation to report employee discipline to the Commission, and undo the ordered undertakings that trigger six years of heightened FINRA supervision. Their argument was that the January 2025 settlements got lighter terms. The Commission denied the motions, over a dissent from Commissioner Peirce, holding that modifying a settled order requires compelling or extraordinary circumstances, and denied a further five-order group in June on the same ground (Release 34-102860, April 14, 2025; 34-103330, June 26, 2025). What you sign in a settlement is what you live with.

The lesson. The examiner’s question here is brutally simple: name every channel your people use to discuss firm business, and show me the archive for each. If there is a channel you cannot produce, the gap between those two lists is the violation, at any firm size. The library’s Electronic Communications & Off-Channel Policy pairs with an Employee Communications Acknowledgement for a reason: the policy sets the channels, and the acknowledgement is the evidence your people knew.

The incident is not the violation. The aftermath is.

In November 2023, ICBC Financial Services was hit with ransomware that encrypted systems and forced it to cut connectivity to its clearing firms. The SEC’s order does not charge the firm for the attack. It charges that for nearly four months afterward, the firm’s books and records were not current and customers did not receive trade confirmations, in violation of Exchange Act Section 17(a) and Rule 17a-3(a), and Rule 10b-10(a) on the confirmations. The order says the quiet part directly: the violations indicate, in part, that the firm “needed to be better prepared for a potentially severe cybersecurity incident” (Release 34-101794, December 2, 2024). Note the outcome, too: because of the firm’s cooperation and remediation, the Commission imposed a censure and cease-and-desist but no civil penalty. The aftermath was the violation, and the response determined the price.

GWFS Equities is the same lesson from a different rulebook. As a retirement plan broker-dealer, GWFS detected escalating account-takeover attempts against plan participants, most of them blocked before money moved. Blocked attempts are still attempted unauthorized access, and GWFS failed to file roughly 130 suspicious activity reports and omitted required information from about 297 more (Release 34-91853, May 12, 2021, $1.5 million, charging Rule 17a-8). Detection without reporting is its own violation.

The lesson. Your recordkeeping obligations do not pause during an outage, and your incident response plan needs a books-and-records continuity workstream, not just a containment checklist. This is also where preparation is most visibly priced: R.T. Jones responded well and paid $75,000; ICBC responded well and paid nothing. An incident response plan that exists before the incident is the whole game, which is why the library’s Incident Response Plan ships with playbooks for credential compromise, ransomware, vendor breach, and BEC, why the IRP Builder will generate one mapped to §248.30 for your specific firm, and why the Incident Response Tabletop Exercise Kit exists to generate that testing evidence. If your plan has never been rehearsed, start with “We Have a Plan” Is Not the Same as “We Tested It”.

One more pattern, about AI claims

Two advisers were charged on the same day in March 2024 for describing their artificial intelligence capabilities more generously than the facts supported. Delphia told clients it used AI and machine learning on client data to inform investment advice; the order finds no such data was used ($225,000, IA-6573). Global Predictions marketed itself as the “first regulated AI financial advisor” and, per its order, could not produce documents to substantiate the claim ($175,000, IA-6574). Both orders charged Marketing Rule violations under Rule 206(4)-1 alongside the antifraud provisions. This is why the method section at the top of this article is worded the way it is. If your firm describes its use of AI, in marketing or in an ADV, the enforcement record treats that description exactly like a performance claim under the Marketing Rule: it has to be true, specifically true, and provable on demand.

The lesson. The examination question is no longer whether you use AI. It is whether what you say about it can be substantiated, and whether the use itself is supervised. The library’s AI Compliance & Operational Framework and AI Governance Workbook exist for the second half of that; for the first, start with Your Firm Is Already Using AI. Here’s How to Make It Defensible.

The request list you can answer today

The record since 2010 reduces to a document request list. Every item below is justified by a named order, not by a framework or a vendor threat report.

  • Written safeguards policies that reach every branch, contractor, and affiliated office, with evidence of adoption, not just distribution (34-104255; 34-92800)
  • MFA enforced and verifiable, not “required wherever possible” (34-92800; 34-84288)
  • A Reg S-ID program built from your own covered accounts and incident history, updated on a schedule, not copied from Appendix A (IA-6073; 34-95369; 34-102170)
  • Vendor oversight and end-of-life media destruction treated as Safeguards Rule obligations, per 17 CFR 248.30 as amended (34-95832)
  • Call-back verification for disbursements and share movements, with proof the procedure is trained and followed, because a warning email nobody confirmed reading was held against the firm (34-100780)
  • An incident response plan with a books-and-records continuity workstream (34-101794)
  • SAR review that treats blocked account takeovers as reportable (34-91853)
  • Compliance review of client breach notifications before they go out; Cetera’s misleading notification timing drew its own Rule 206(4)-7 charge (34-92800)
  • Complete capture of business communications, on every channel your people actually use (the off-channel program above)
  • Any material statement of fact about your firm’s use of AI in an advertisement held to the Marketing Rule’s substantiation standard, Rule 206(4)-1 (IA-6573; IA-6574)

If you want to know how many of those ten you could produce this afternoon, the Mock SEC Cyber Exam will tell you in about seven minutes, and it is free.

The SEC has never needed a cybersecurity rule to bring these cases, and it does not examine your firewall. It examines whether the program your documents describe is the program your firm actually runs. 37,299 documents, one sentence.

The BrainTrust, free

This article is public. A free BrainTrust account opens every Insight, the 13 Sample Deliverables, the full Mock SEC Cyber Exam and the core compliance briefs. Name and email, no card.

Create Free Account →
Outside Testing

Want an outside test of your program? One-time engagements are available only to BrainTrust Premium members →