Digital Evidence in the Courtroom: A 2026 Mid-Year Review

For decades, the elements of attorney-client privilege have been stable: a communication made in confidence, between privileged persons, for the purpose of obtaining or providing legal assistance for the client. In the first half of 2026, a cluster of federal rulings tested how those familiar elements apply when the "communication" is a prompt typed into a generative AI platform — and the courts did not agree on the answer. Between February and May 2026, four federal courts issued rulings that collectively define the first contours of an AI-use governance framework for discovery, and they do not all point in the same direction. For general counsel and litigation partners, the resulting fracture is not an academic curiosity awaiting appellate cleanup. It is the operative reality for every legal team that has fed a confidential document into an AI tool and assumed the output was beyond opposing counsel's reach.

Two Courts, One Question, Opposite Answers

The fault line opened in United States v. Heppner (S.D.N.Y., Feb. 17, 2026), where Judge Jed Rakoff held that litigation preparation materials a criminal defendant generated through a public consumer AI platform were neither attorney-client privileged nor protected work product. The reasoning was structural: the platform's terms of service explicitly permitted user data to be used for model training, which destroyed any reasonable expectation of confidentiality — and without that expectation, the foundational element of privilege did not exist.

One week earlier, Warner v. Gilbarco, Inc. (E.D. Mich., Feb. 10, 2026) reached the opposite conclusion in a civil matter, reasoning that AI platforms are "tools, not persons," and that routing work product through an AI system is not disclosure to an adversary in the legally cognizable sense required to waive protection. The doctrinal question dividing the two courts is genuine: does the privilege analysis turn on the nature of the receiving party — human adversary versus automated system — or on the contractual and technical realities governing what happens to the data once submitted? Heppner answers with the latter; Warner with the former.

The synthesis arrived in Morgan v. V2X (D. Colo., Mar. 30, 2026), where a magistrate judge upheld work product protection for a pro se plaintiff's AI-assisted litigation materials but attached two conditions: the plaintiff had to disclose the name of any AI tool used with confidential materials, and the court amended the protective order to prohibit consumer-tier AI on any confidential ESI unless the vendor contractually bars training on submitted data, restricts onward disclosure, and permits deletion on demand. That triad — training prohibition, disclosure restriction, deletion right — now functions as a practical compliance checklist, and consumer-tier versions of ChatGPT, Claude, and Gemini will presumptively fail it absent specific enterprise agreements. The through-line is unmistakable: the privilege analysis now begins not with what was communicated but with where it was communicated, and counsel must treat AI platform selection as a threshold privilege decision, not an IT procurement matter.

The Expert's Prompts Are Part of the Method

The fracture reached expert practice on May 18, 2026, when Connecticut Magistrate Judge Thomas O. Farrish ruled in Conservation Law Foundation v. Shell Oil, No. 3:21-cv-00933 (D. Conn.), that AI prompts an expert used to cull documents before forming an opinion are discoverable under Rule 26. The expert's proponents argued that the parties' Rule 29 stipulation shielded the prompts; the court rejected that on two independent grounds — the agreement was ambiguous as to AI-generated materials, and ambiguity in a discovery-limiting agreement cannot exclude otherwise discoverable material, while the prompts themselves were functionally inseparable from the expert's analytical process under Rule 26(a)(2)(B) and 26(b)(4). It is the first federal ruling to treat AI-assisted document culling as a component of expert methodology subject to Rule 26 disclosure, though the order has been stayed pending the district court's review of a Rule 72(a) objection.

The forensic logic is sound. When an expert delegates initial document identification to a large language model, the prompt becomes the first filter between the full evidentiary record and the expert's analysis; an expert who prompts a model to find documents "supporting" a conclusion will receive a different set than one who asks for all documents "relevant to" it — a methodologically significant difference that is now discoverable.

Sanctions for the Copy-Paste Lawyer

A further data point concerns counsel's own conduct. In White v. Walmart, Inc., No. 1:25-cv-01120 (S.D. Ind. Apr. 14, 2026), Magistrate Judge Tim A. Baker sanctioned plaintiff's counsel who had uploaded the defendant's discovery responses into an AI program, asked it to identify insufficient responses, and copied the results into a meet-and-confer email sent to opposing counsel and the court — holding that exclusive reliance on AI output, without independent legal analysis, does not satisfy Rule 26's good-faith conferral requirement. Rule 26(g) requires an attorney's signature certifying, after reasonable inquiry, that a discovery submission is consistent with the rules; when counsel substitutes machine output for that judgment, the certification is violated in substance even if satisfied in form. Courts are not hostile to AI-assisted discovery work — what they are hostile to is AI used as a liability shield, and they are now drawing that distinction explicitly.

What This Means for In-House Teams

Read together, the 2026 rulings establish that every stage of AI-assisted litigation leaves an evidentiary footprint: prompts, outputs, and full conversation logs preserved by the platform may all constitute ESI subject to discovery obligations, depending on the jurisdiction, the platform, and how clearly the parties addressed the issue in their ESI protocol. Most Rule 26(f) conferences do not yet address AI-generated ESI with any precision, and that gap is becoming dangerous.

The response is governance, not abstinence. Organizations should differentiate consumer and enterprise AI tools in written retention and privilege policies, issue AI usage guidance specifying which platforms are approved for matter work, and make AI-generated ESI an explicit topic at every Rule 26(f) conference. Retaining counsel should ask testifying experts at engagement — not on the eve of disclosure — whether they use any AI tool for document review, research, or analysis, and should preserve prompt logs, model versions, and configuration metadata from the outset. Every AI output that informs a filing or meet-and-confer communication should be reviewed by a licensed attorney, with that review documented contemporaneously rather than reconstructed after a sanctions motion.

None of this is optional under the professional rules. Comment 8 to ABA Model Rule 1.1 requires lawyers to keep abreast of the benefits and risks of relevant technology, and reliance on a vendor does not transfer that responsibility. Attorneys have been sanctioned for inappropriate use of ChatGPT, and some judges have demanded that counsel attest to not having used the technology at all. The lesson of early 2026 is that prompt hygiene and platform selection are now privilege decisions with case-dispositive consequences — and the parties who document a defensible AI governance posture before discovery opens will be the ones who never have to litigate it.

For years, a familiar script played out in discovery disputes: relevant text messages were gone, counsel offered explanations involving phone upgrades or app defaults, and courts — skeptical but cautious — hesitated to treat consumer-grade auto-deletion as the equivalent of deliberate document shredding. That era is over. A converging body of judicial decisions, regulatory enforcement actions, and agency guidance issued between 2024 and 2026 has hardened a new premise: when a party knows, or reasonably should know, that litigation is coming and allows ephemeral messaging features to keep running, the resulting data loss is a choice — and courts are sanctioning it accordingly.

The Judicial Record Has Caught Up to the Technology

Two decisions mark how far the judicial posture has moved. In Oakley v. MSG Networks (S.D.N.Y. July 2025), Judge Sullivan imposed spoliation sanctions after AT&T call-detail records showed the plaintiff had sent 1,113 text messages in the three weeks immediately following the incident at issue — a window in which preservation obligations had plainly attached — and none were preserved; the plaintiff had also cycled through seven phone upgrades over the course of the litigation. The court's use of carrier-level metadata as the baseline for measuring preservation failure signals a new evidentiary toolkit for spoliation motions: the messages may be gone, but the record of their existence is not.

The Ninth Circuit went further. As analyzed in Paul, Weiss's April 2, 2026 column in the New York Law Journal, the court affirmed dismissal with prejudice in an employment case where the plaintiff and key witnesses deleted relevant text messages and defied multiple court orders to preserve and produce them. Dismissal with prejudice confirms that text-message spoliation is no longer treated as a lesser infraction than the destruction of email or formal documents.

Counsel should also keep the underlying storage concepts in view. Unallocated space — the area of a drive available to store new files — may still contain old data until that data is overwritten by newer data, which is why deleted files and unallocated space remain basic concepts every litigator handling electronic evidence needs to understand.

Regulators Priced In the Risk — Then Washington Went Quiet

The regulatory record set the stage. The Securities and Exchange Commission has collected more than $2.8 billion in fines tied to off-channel communication failures across dozens of financial institutions whose employees routed business conversations through WhatsApp, Signal, and personal email, with the enforcement posture extending beyond broker-dealers to investment advisers and private equity firms. The DOJ and FTC's January 2024 joint guidance expressly named Slack, Signal, and collaborative chat platforms as covered by records-retention and antitrust-investigation obligations. That guidance acquired immediate forensic significance: in the DOJ's Google Search monopolization case, prosecutors presented evidence that Google had set Google Chat history to "off" by default and trained employees on deletion practices, and in FTC v. Amazon, senior executives reportedly continued using Signal's disappearing-message feature for competition-related discussions after retention obligations were known.

Then, beginning in January 2025, the Trump Administration effectively paused SEC and CFTC enforcement of off-channel and ephemeral messaging rules. Some compliance departments exhaled. They should not have. As Quinn Emanuel's April 2026 client alert put it with unusual directness, companies interpreting the pause as license to loosen messaging governance are making a dangerous mistake: state attorneys general, private litigants, bankruptcy trustees, class-action plaintiffs, and foreign regulators operate on entirely independent tracks, and none of them paused anything.

The civil standard has not moved. Federal Rule of Civil Procedure 37(e) asks whether a party with a duty to preserve took reasonable steps to prevent loss of relevant ESI — not whether the SEC is active — and sanctions range from adverse inferences to case-dispositive relief with no regulatory predicate whatsoever. The Albertsons litigation is illustrative: litigation holds were technically issued, but custodians never disabled auto-delete on messaging applications, so the hold existed on paper while evidence evaporated in fact. International exposure is escalating on its own track as well: in June 2024, the European Commission fined International Flavors & Fragrances, jointly with its French subsidiary, €15.9 million after an employee deleted WhatsApp messages relevant to a competition investigation, and EU regulators have signaled the action is not isolated.

The Hold That Does Not Name the Platform Is Not a Hold

The structural vulnerability is behavioral, not doctrinal. A 2025 Global Relay survey found that more than 70% of financial services firms report employees using personal or ephemeral applications for business purposes despite explicit written prohibitions — a figure PlatinumIDS echoed in April 2026. Policy alone has demonstrably failed to change behavior, and the 2026 communication environment — Teams, Slack, WhatsApp, Signal, iMessage, each with different retention defaults and API architectures — has widened the gap between issuing a hold and actually preserving data.

The prescription is specific. Every litigation hold issued today must name ephemeral platforms explicitly — Signal, WhatsApp, Google Chat, iMessage — rather than relying on a generic instruction to "preserve relevant communications"; must direct recipients to disable auto-delete and disappearing-message features at the device and application level with date-stamped confirmation; and, where BYOD devices are in scope, must address whether IT can technically reach them. Modern data forms belong in the same instrument. Cloud hyperlinks — the "modern attachments" that replace static files with pointers to living documents in Google Drive, SharePoint, or Box — break the traditional parent-child collection model and present a point-in-time versioning problem, and the prevailing trend among courts and parties is to address them expressly in Rule 26(f) conferences and ESI protocols rather than discover the gap at production. Ambiguity left unresolved at the protocol stage becomes a dispute, and potentially a sanctions risk, later.

These are ethics questions as much as discovery questions: eDiscovery missteps implicate counsel's legal-ethics obligations and carry sanctions exposure of their own. And when preserved messaging data is withheld as privileged, FRCP 26(b)(5)(A)(ii) still requires a description sufficient for the opposing party to assess the claim — meaning chat and messaging content must be accounted for in the privilege-log workflow, not quietly omitted. The organizations that treat the federal pause as a window to harden technical controls will be positioned for the next civil discovery request, state investigation, or EU inquiry; those that treat it as permission to stand down will not.

For decades, a photograph, a recording, or a document carried a working presumption that it depicted something that actually happened. That presumption is no longer safe. Synthetic media can now fabricate a face, a voice, or a signature with a fidelity that defeats the naked eye and the untrained ear. Deepfakes are produced by generative adversarial networks — two AI systems pitted against each other, one generating synthetic media and one attempting to detect it, until the forgery becomes indistinguishable from the genuine article. The result is a class of fabrication that can replicate a specific person's face, voice cadence, and mannerisms, and that is arriving not in law-review hypotheticals but in dockets and deal rooms today.

The Fabrications Are Already Here

The commercial world absorbed the first losses. In 2019, a UK energy company lost roughly $243,000 when attackers used AI to clone the voice of its parent company's chief executive and authorize a fraudulent wire transfer — the first widely reported case of its kind. In a Dubai bank fraud revealed in 2021, criminals used the same voice-cloning technique to move $35 million. The trend lines are steep: the Deloitte Center for Financial Services projects that generative AI could push U.S. fraud losses from $12.3 billion in 2023 to $40 billion by 2027, a 32 percent compound annual growth rate. A Regula study found that 50 percent of businesses reported encounters with video deepfakes, up from 20 percent in 2022; across industries, businesses have lost an average of nearly $450,000 to deepfakes, with financial-services losses exceeding $600,000 on average. In an Ironscales survey, 75 percent of organizations reported at least one deepfake-related incident in the prior year, and 60 percent described themselves as only somewhat confident, or not confident at all, in their ability to detect the next one.

What begins as fraud ends as evidence. Family court illustrates the stakes: in a contested custody matter, a parent can fabricate videos depicting the other parent verbally abusing a child — recordings indistinguishable from genuine footage to the untrained eye — and the accused parent must then undertake the complex and costly process of retaining digital-forensics experts to prove the videos are fakes. Arbitration amplifies the exposure: as ADR has grown in commercial and employment disputes, its historically lighter evidentiary gatekeeping makes it an attractive venue for bad-faith evidence submission.

Where the Rules Strain

The Federal Rules of Evidence still supply the framework, but the bar for satisfying them has moved. Rule 901 requires evidence "sufficient to support a finding that the item is what the proponent claims it is" — a standard designed to screen out accidental or amateur manipulation, not AI systems trained on thousands of hours of footage to produce a forensically indistinguishable clone. A sponsoring witness who testifies "this is a fair and accurate depiction" may be entirely sincere and entirely wrong, because a convincing fake can deceive the very people offering it. Rule 901(b)(9) permits authentication of a process "shown to produce an accurate result," but no consensus standard yet exists for deepfake detection. Rule 403's unfair-prejudice screen falters for a different reason: the trier of fact cannot calibrate prejudice from a fabrication sophisticated enough to appear genuine without expert assistance — the jury cannot discount what it cannot identify.

Rule 902's self-authentication provisions offer only partial help. Subsections (13) and (14), covering certified machine-generated records and data copied from electronic devices, authenticate the integrity of a file as captured; they do not vouch for whether the underlying content was synthetically generated before it ever reached the device. That gap is where disputes now live.

The Liar's Dividend — and Hallucinated Metadata

The deepfake era cuts in both directions. Genuine evidence now faces reflexive "it could be fake" challenges, and courts have begun drawing the sensible line: an opponent cannot defeat admissibility merely by invoking the abstract possibility of deepfakes; a concrete basis to question the specific exhibit is required before a heightened burden shifts to the proponent. Counsel defending authentic media must be prepared to preserve originals, document chain of custody from capture forward, and explain ordinary recording artifacts before an adversary spins them as signs of forgery.

A quieter risk compounds the problem: generative AI can corrupt the very metadata lawyers rely on to authenticate documents. Ask a generative tool to clean up, summarize, or reformat a file, and many systems produce a new artifact with freshly minted properties — author names, creation dates, application signatures — that bear no relationship to the original. AI-assisted review platforms and conversion utilities can strip or rewrite last-modified dates, custodian information, EXIF data, and tracked-changes history as they re-save files, leaving visible content identical while the evidentiary timeline silently falls apart. Routing originals through a tool that re-saves them breaks hash values, severing chain of custody and inviting spoliation arguments — a file can be substantively accurate yet evidentiarily worthless. The ethical stakes track the evidentiary ones: the duties of competence, candor, and supervision now reach the AI features embedded in review and processing workflows counsel never inspected.

The Answer Is Forensic, Not Rhetorical

Authenticity disputes are won with methodology, not adjectives. Four disciplines are converging into the price of admission for digital exhibits.

Provenance and metadata analysis. Where media came from is often more revealing than how it looks. Original files carry creation timestamps, device identifiers, and codec signatures; discrepancies with the surrounding facts — or the absence of expected metadata — are red flags worth running down. Content-credential standards such as C2PA attach cryptographically signed records of a file's capture and editing history, and counsel should demand native originals rather than re-exported copies that strip that chain.

Chain-of-custody discipline. Baseline protocols belong at intake, not at rebuttal: custody documentation from the moment of collection, cryptographic hashing to fingerprint the original, and metadata preservation recording device identifiers and file provenance. Evidentiary files should be quarantined from any generative pipeline that re-saves or "enhances" them, with AI output treated as a working copy, never the evidentiary original.

Detection tooling, honestly framed. Examiners look for convergence across independent indicators — lighting and shadow inconsistencies, unnatural blinking, compression and error-level anomalies in imagery; spectral patterns, breath sounds, and speech micro-timing in audio — documented so another examiner can reproduce the analysis. No single artifact is dispositive; responsible findings are expressed in terms of consistency and probability, not false certainty.

Qualified expert testimony. When authenticity turns on technical analysis, Daubert gatekeeping governs: testable methods, known error rates, peer review, general acceptance. A qualified examiner verifies file integrity, reconstructs custody, interrogates metadata and compression artifacts — and explains in plain terms what the evidence does and does not establish, so the factfinder is neither lulled into false confidence nor frightened into rejecting reliable proof.

Counsel in complex commercial matters should go one step further and negotiate deepfake-authentication provisions into discovery protocols and ESI agreements, specifying the technical standards against which contested media will be measured; courts have shown willingness to appoint forensic neutrals for precisely this kind of technically complex evidence issue. The structural stake is not one exhibit or one verdict. If courts cannot reliably distinguish genuine from fabricated digital evidence, the adversarial system's foundational promise — that truth can be found through the presentation and testing of evidence — is undermined at its root.

Commercial litigation suffers from a recurring, expensive dysfunction: two sophisticated parties each retain the most credentialed technical experts their budgets allow, those experts reach diametrically opposed conclusions from the same evidence, and the court — lacking independent technical competence to resolve the conflict — is left to make credibility determinations among witnesses whose primary function is advocacy, not analysis. In 2026, the pressures on that model have become impossible to ignore. The sheer volume and diversity of digital data and the technical intricacies of managing and retrieving it, together with the consolidation of hundreds of related actions into single proceedings, have pushed courts and counsel toward a structurally different answer: a single qualified neutral who serves the proceeding rather than either litigant.

The Model: One Expert, Serving the Court

A forensic neutral is appointed by a court, mutually retained by the parties, or designated by an arbitrator to perform technical work in service of the proceeding itself. The role is genuinely dual: forensic neutrals are typically qualified both as attorneys and as technologists, allowing them to operate at the intersection of legal process and technical analysis. Federal Rule of Civil Procedure 53(a)(1)(A) supplies the foundational authority in federal litigation, empowering courts to appoint a neutral to perform duties to which the parties consent, to address matters that cannot be efficiently handled by the district judge, and to perform accountings or complex technical computations without party consent in appropriate circumstances.

The practical mandate is broad: drafting and auditing compliance with forensic collection and processing protocols, performing independent forensic analysis of electronic evidence (including deleted or fragmented data), identifying digital artifacts that establish access, copying, or transmission events, validating or challenging the methodologies of each party's experts, and translating highly technical findings into terms the trier of fact can assess. Trade secret disputes illustrate the value with particular clarity. A finding of misappropriation is substantially easier to obtain than proof of complete remediation, and a neutral operating under court authority is the only party positioned to oversee court-ordered data purging and certify completion credibly — a capability the 2016 Defend Trade Secrets Act made more consequential by providing for ex parte seizure orders whose execution demands highly technical identification, collection, and deletion of specific data from specific systems.

The economics reinforce the logic. Two retained experts do not produce twice the analytical quality; they produce twice the cost, extended discovery timelines, extended briefing cycles, and an outcome that turns on credibility assessments rather than technical accuracy. A neutral eliminates the duplication: costs are shared, the analysis is performed once, and the court is not asked to adjudicate a secondary conflict about methodology — an efficiency that is transformative in disputes where technical expert fees routinely reach seven figures and discovery timelines are measured in years.

Where the Trend Is Playing Out: E-Discovery, Social Media MDLs, and Mass Torts

The same logic now extends across the discovery process itself. Technical e-discovery neutrals — legal professionals with expertise in both law and technology — advise parties on technical issues, ensure compliance with relevant legal standards, and arbitrate disputes over ESI, in roles ranging from consultative engagements to formal court appointments in complex litigation. Their involvement can narrow disputes and save significant time and cost in a landscape where the sheer volume and diversity of digital data has made discovery a technical discipline in its own right.

Nowhere is the need more visible than in consolidated proceedings. Approximately 457,000 civil actions were pending on MDL dockets as of December 31, 2023 — roughly 67% of all pending civil litigation in the United States. MDLs consolidate cases from disparate jurisdictions for pre-trial proceedings, including discovery, case management, Daubert determinations, and dispositive motions, and they often involve large volumes of data consolidated from disparate sources.

The wave of social media litigation shows what that means in practice. Hundreds of cases alleging potentially life-threatening effects of adolescent social media use have been filed in state and federal courts, many consolidated into MDLs or within state court departments — including a California state court action in which the court overruled Snap's demurrer on claims that Snap's conduct in designing and implementing Snapchat resulted in the foreseeable deaths of plaintiffs' children, who overdosed on fentanyl. Social media repositories present unique discovery problems: the data is controlled by third parties such as Meta, X, or Snap, and a user's download captures only a snapshot at a moment in time, missing any subsequent changes. A discovery special master can streamline these cases and ensure the right data is collected as efficiently as possible — expertise that lawyers and judges, who may be unfamiliar with these platforms and the unique types of data they generate and store, often lack.

The neutral's role does not end when liability resolves. In mass torts, courts increasingly appoint neutral special masters to administer common benefit funds — the holdback-financed pools that compensate the firms whose work advanced the entire litigation. The structural problem is stark: lead counsel who built the litigation are also claimants on the fund, allocations often involve tens or hundreds of millions of dollars, and even a scrupulously fair internally negotiated split invites appeals, fee disputes, and reputational damage. Rule 53 requires the court to address the special master's potential grounds for disqualification before appointment, and a well-run allocation rests on published criteria, consistent treatment across firms, and a genuine opportunity to contest preliminary findings — process features that give the result a legitimacy no self-interested division can match.

The AI Frontier — and the Question for Counsel

The institutional infrastructure is formalizing. In April 2024, JAMS announced new rules for disputes involving artificial intelligence, covering the protection of proprietary training data and AI models and the knowledge needed to arbitrate disputes concerning AI software. The rules define AI as "a machine-based system capable of completing tasks that would otherwise require cognition" — no small contribution in an industry where judges have differed on what the term even entails. JAMS President Kimberly Taylor explained that the rules anticipate new litigation arising from the technology and reflect a judgment that resolving AI-driven disputes via arbitration will require tailored rules accounting for the technology's complexity and the novel factual and evidentiary issues it introduces. The message for counsel is that the neutral model is no longer improvised case by case; a major ADR institution has now built rules around it.

For general counsel and litigation partners, the strategic calculus has accordingly shifted. The threshold question is whether technical complexity is driving costs and delays beyond what the case value justifies, or distorting the proceedings in ways that compromise fair resolution — and if the answer to either is yes, a neutral belongs on the table. The clearest triggers: experts at fundamental methodological odds, court-ordered collection or deletion that no partisan expert can credibly oversee, and arbitrations and mediations where the absence of formal evidentiary rules makes a technically qualified neutral an even more valuable objective anchor. Counsel should evaluate a forensic neutral appointment at the outset of litigation, not after discovery has produced an intractable battle of experts: early appointment consistently reduces costs, shortens timelines, and produces more defensible outcomes. Courts are increasingly receptive to such appointments; the question is whether counsel asks.

If the first half of this decade taught security executives that breaches are inevitable, the first half of 2026 confirms a harder lesson: enforcement risk has become personal, and it has become documented. The individuals who shape a company's security messaging — and the directors who attest to overseeing its program — now sit inside a chain of representations that runs from the security operations floor into public filings. When any link in that chain diverges from documented reality, the exposure is no longer abstract or corporate. It attaches to people.

From the Engine Room to the Caption: Sullivan and SolarWinds

Two enforcement actions did the work of reshaping executive liability. The criminal conviction of Uber's former security chief, Joe Sullivan, in connection with the company's handling of a 2016 data breach, established that a security leader can be held personally accountable for how a breach is managed and characterized after the fact — the response, not the intrusion, carried the legal consequence. Then, on October 30, 2023, the SEC sued SolarWinds and its CISO, Timothy G. Brown, alleging that from October 2018, when SolarWinds went public, through January 2021, the company and Brown defrauded investors by overstating SolarWinds' cybersecurity practices and understating or failing to disclose known risks — disclosing only generic and hypothetical risks at a time when, the SEC alleges, both knew of specific deficiencies in the company's security practices. The underlying SUNBURST attack, attributed to Russian state-sponsored actors, affected over 18,000 customers, including government agencies and Fortune 500 companies. The Wells Notice that preceded the suit, received by current and former SolarWinds executives in June 2023, was the first instance of a CISO being directly named.

Read together, these matters crystallize a single principle: regulators and prosecutors are willing to look past the corporate entity to the individuals who shaped security messaging and disclosure decisions, and misalignment between representation and reality is the exposure. For counsel, the operational consequence is that documentation is now a form of defense — risk assessments, remediation decisions, accepted-risk acknowledgments, and incident timelines must be created contemporaneously and written with the assumption that a regulator or jury may one day read them, while optimistic internal characterizations of known weaknesses can later be contrasted with external assurances to damaging effect. CISOs should also press boards and general counsel on directors-and-officers coverage, indemnification commitments, and access to independent counsel where the individual's interests and the company's may diverge.

The Attestation Chain in Public Filings

The SEC's 2023 cybersecurity disclosure rules converted oversight from an internal management concern into a documented, externally disclosed chain of accountability. Form 8-K Item 1.05 requires disclosure of a material cybersecurity incident within four business days of determining that it is material — the clock runs from the materiality determination, not from discovery, which makes the process for reaching that determination a governance artifact in its own right. Regulation S-K Item 106 requires annual disclosure of the company's cybersecurity risk-management strategy and governance, including a description of the board's oversight of cybersecurity risk.

Item 106 is what forges the chain. The security team's representations inform the board's understanding; the board's understanding is memorialized in its oversight description; that description becomes a public disclosure investors rely on — and when the described posture diverges from documented reality, every link is exposed to scrutiny at once. Boards cannot validate a program by accepting management's self-assessment of management's own work, which is why independent audits have moved from best practice to near-necessity; New York's Department of Financial Services makes the point explicit, requiring covered entities to conduct independent audits of their cybersecurity programs alongside more frequent risk assessments, ransomware-payment reporting, and annual training. Disclosure also cuts against a competing duty: corporate leaders must weigh the new reporting requirements against the risk that disclosures expose trade secrets, a tension boards must resolve deliberately rather than by default.

The obligations no longer stop at large public companies. The amended Regulation S-P requires broker-dealers and registered investment advisers — regardless of size — to develop, implement, and maintain a written incident-response program designed to detect, respond to, and recover from unauthorized access to or use of customer information, and imposes an affirmative duty to notify affected individuals when sensitive customer information has been, or is reasonably likely to have been, accessed or used without authorization. The rule also makes vendor risk the firm's own: firms must oversee service providers through contractual arrangements ensuring prompt breach notification, an area where standard vendor agreements frequently fall short.

The Ransomware Overlay: Sanctions on Top of Disclosure

Ransomware compresses all of these duties into hours and layers federal sanctions law on top. Under the OFAC framework — including the Treasury Department's October 2020 advisory — payments to designated threat actors, including groups affiliated with North Korea, Iran, and Russia, are prohibited regardless of intent, and the Specially Designated Nationals list includes ransomware collectives such as Evil Corp, identified as operating under Russian state direction. Because OFAC's civil enforcement is strict-liability in character, a company that pays even unknowingly and in good faith can face civil penalties, which makes pre-payment due diligence — threat-actor attribution, cross-referencing against OFAC's list, consultation with counsel — a legal requirement, not a courtesy. Meanwhile the four-business-day 8-K clock and state breach-notification statutes — now on the books in all 50 states, some with deadlines as short as 30 days from discovery — run in parallel, and cyber insurance adds conditions of its own: some policies require insurer approval before payment, war exclusions were actively litigated after NotPetya, and some policies now condition ransomware coverage on OFAC due diligence having been performed.

Pressure That Does Not Pause

Counsel should not assume that any softening in one enforcement channel lowers the temperature overall. GDPR fines for non-compliance have continued to increase, 2023 brought the EU's move toward a more centralized enforcement model allowing greater consistency across member-state rulings, and the largest GDPR fine to date was issued against Meta. State regimes — from the 50-state breach-notification patchwork to NYDFS's audit and reporting mandates — supply independent pressure whatever the federal posture in a given year.

The through-line for general counsel is a documented, defensible playbook built before the incident: a materiality-determination process owned jointly by security, legal, finance, and disclosure committees; a genuinely independent audit whose findings are escalated, tracked, and remediated; a tabletop exercise that sequences OFAC diligence, disclosure timing, notification clocks, and insurance conditions with legal, IT, finance, and communications in the same room; and a reconciliation of the company's public security representations with its documented reality before the next filing, not after the next incident.

If the decisions and enforcement actions surveyed in this report share a single through-line, it is this: legal exposure now attaches to what an organization can prove, not what it intended. When an AI system produces a discriminatory outcome or a legal error, the question regulators and plaintiffs ask is not "did you intend this?" but "did you have a governance framework, and did you follow it?" When relevant messages disappear from an executive's phone, courts and regulators no longer accept platform defaults as exculpatory — the question is whether the organization took affirmative steps to preserve, and if the answer is no, the inference runs against it. And when a vendor's failure exposes customer data, "we didn't know" does not satisfy an exam team that can point to six months of unreviewed vendor security questionnaires. The practical posture for the next eighteen months is forensic readiness: building, before the crisis, the documented record that intent alone no longer supplies.

AI: A Liability Multiplier Without a Framework

AI deployment is creating new categories of legal risk faster than most governance structures can absorb them. Every deployed system generates documented decisions, a traceable algorithm, and a data trail that regulators, opposing counsel, and class action plaintiffs will eventually examine — and the FTC, CFPB, EEOC, and SEC have all demonstrated willingness to scrutinize algorithmic decision-making and hold organizations accountable for discriminatory or misleading outcomes regardless of intent. The predictable failure modes are already visible: credit models incorporating proxy variables that create fair lending liability across millions of decisions, hiring algorithms that bake in historical bias, and document review systems whose misclassification of privileged communications produces sanctions exposure in litigation.

For counsel, the obligation is personal as well as institutional. Comment 8 to ABA Model Rule 1.1 makes competence with relevant technology an ethical duty, and AI-assisted research tools have already produced fabricated citations in filed briefs — the tool does not bear the professional responsibility; the attorney does. The governance answer rests on four documented pillars: defined objectives and ethical principles reviewed regularly as models drift, named accountability including shutdown authority, training-data provenance documentation, and auditability sufficient to reconstruct the system's state at any historical decision point. GCs procuring AI tools should also apply the privilege lens: privilege protects only communications made in confidence, between privileged persons, for the purpose of legal assistance — a standard that argues for enterprise-tier deployments and contract terms that preserve confidentiality, particularly given that recent rulings such as Garner v. Amazon.com Inc. (W.D. Wash., September 2024) and Epic Games Inc. v. Apple Inc. (N.D. Cal., December ruling) show how contested in-house privilege assertions have become.

Vendors: Still the Weakest Link in the Evidentiary Chain

The 2013 Target breach — executed through an HVAC contractor, not Target's own perimeter — remains the paradigm, and the lesson still has not fully penetrated corporate governance: the organization that hired the vendor bears most of the legal consequences when the vendor is exploited. Regulators have converted that lesson into direct obligation. The New York DFS cybersecurity regulation requires covered entities to assess third-party providers' cybersecurity practices and embed specific requirements in vendor contracts, with parallel obligations under SEC supply chain guidance and FFIEC guidance for financial institutions.

The contract is the primary instrument, and standard vendor paper is drafted to minimize vendor liability. Counsel should insist on measurable security standards tied to a recognized framework (NIST CSF, ISO 27001, SOC 2) rather than "industry standard" language; breach notification within 24 to 72 hours; independent audit rights; cyber insurance naming the organization as additional insured; and indemnification tied to specific security failures. Critically for forensic readiness, contracts should guarantee the primary organization's security and legal teams access to the vendor's incident investigation — learning the scope of an incident involving your customer data from your vendor's press release creates notification timing exposure and forecloses coordinated remediation. With legal fees, notification costs, and settlement exposure in a significant vendor-caused breach routinely reaching eight figures, the program cost is an insurance premium, not overhead.

Boards: Documented Oversight, Independently Verified

Board members increasingly face personal liability for their organization's cyber posture. The SEC's 2023 rules require boards to oversee cybersecurity management and demonstrate active oversight, require disclosure of material incidents within four business days, and require description of the board's oversight role and risk-management processes. The amended NYDFS regulation goes further, requiring covered entities to conduct independent audits of their cybersecurity programs and to report ransomware payments. Independent cybersecurity audits are the cleanest mechanism for discharging these duties: they demonstrate proactive leadership, reveal blind spots, and supply the dynamic evidence of compliance that regulators now demand.

The Forensic-Readiness Checklist for the Next 18 Months

The affirmative agenda for counsel follows directly from the enforcement record.

Reissue litigation holds that name the technology. After Oakley v. MSG Networks — where carrier records showing 1,113 unpreserved post-incident texts and seven phone upgrades drew spoliation sanctions — and a Ninth Circuit affirmance of dismissal with prejudice for deleted texts and defied preservation orders, generic hold language is indefensible. Holds must name Signal, WhatsApp, Google Chat, and iMessage explicitly, direct recipients to disable auto-delete with date-stamped confirmation, and confront the BYOD gap that a 2025 Global Relay survey quantified: more than 70% of financial services firms report employees using personal or ephemeral apps for business despite written prohibitions. The SEC's more than $2.8 billion in off-channel fines and the DOJ/FTC January 2024 joint guidance naming Slack and Signal confirm the regulatory floor.

Pre-identify forensic neutrals. Where electronic evidence is central, a court-appointed neutral under Federal Rule of Civil Procedure 53(a)(1)(A) can draft and audit collection protocols, validate competing methodologies, certify remediation that no partisan expert can credibly provide, and execute Defend Trade Secrets Act ex parte seizure mandates — at shared cost and on shorter timelines than a battle of experts. Early appointment, evaluated at the outset of litigation rather than after expert deadlock, consistently reduces cost and produces more defensible outcomes.

Build the incident playbook before the demand arrives. A ransomware event triggers overlapping clocks: OFAC due diligence before any payment — civil exposure for paying a designated entity such as Evil Corp is effectively strict liability, even after good-faith diligence — the SEC's four-business-day Form 8-K window running from the materiality determination, and state notification deadlines as short as 30 days. Counsel should be designated an immediate stakeholder so privilege attaches to the investigation from day one, insurance policies should be read cover to cover for pre-payment approval and OFAC-diligence conditions and war exclusions litigated after NotPetya, and the full sequence should be run in a tabletop exercise — where process gaps are found in simulation rather than in the event itself.

The organizations that emerge intact from 2026–27 will not be the ones that avoided every incident. They will be the ones that can produce, on demand, the governance framework, the vendor file, the audit report, and the preservation record — proof, assembled before anyone asked for it.