By Joanna Fields is the founder and CEO, Aplomb Strategies

In honor of Halloween, I was recently asked for a good horror movie recommendation, which got me thinking: does anyone remember The Ring? The 2002 horror classic started with an innocent act: someone pressed play on an unmarked videotape. Does anyone remember videotapes? The moment it was watched, the damage was done, the phone rang, and the clock started ticking. The only way to escape the curse was to copy the tape and pass it on, spreading it further with every attempt to contain it.

More than twenty years later, the cursed tape may have a modern sequel: unsecured AI tools that could start a chain of events with ominous outcomes – kicked-off by an unaware agent whose seemingly benign actions spin beyond their control. What if:

➢ A trader under investigation types privileged details of his case into a public chatbot to assess his own exposure.

➢ An engineer uses AI to rewrite model parameters in a database outside the secured and controlled system.

➢ An investment banker joins a confidential call about a client’s unannounced acquisition while the firm’s AI assistant records, transcribes, and summarizes it.

In each case, a single keystroke could send confidential information beyond the organization’s control and trigger a series of seemingly inevitable consequences.

The depth of the nightmare lies partially in the AI tool’s terms, including: what data can be stored, reviewed by third parties, or used to train future models. As soon as it has been “seen” by AI – it can’t be “unwatched.” A series of events could be initiated where terms like, “privilege, confidentiality or privacy” may be argued to be waived, trade secrets may lose protection, or novelty for a future patent may be compromised. Like the cursed tape in The Ring, the understated consequences have heavy implications and arrive on their own schedule.

Heroic Attempt to Break the Curse from Detroit

Just when every AI chat log seemed like a cursed tape bound for discovery, a Detroit court recently offered a plot twist. In Warner v. Gilbarco, Inc. (E.D. Mich. Feb. 10, 2026), Magistrate Judge Anthony P. Patti held that a self-represented plaintiff’s use of ChatGPT did not waive work product protection. AI, he reasoned, is a tool, not a person, and waiver requires disclosure to an adversary.

But like any good horror sequence, the relief was short-lived. One week later, Judge Jed Rakoff in New York rejected privilege protection for a criminal defendant’s AI communications. Because Warner limited the ruling by addressing only work products, it offers little shelter for privileged communications, trademark strategy, or unfiled patents.

With courts splitting outcome decisions, individual states writing their own AI and privacy rules, and regulators racing to catch up – there is no single map for protecting confidential, privileged or proprietary information in the U.S. The rules change depending on where the tape is played.

The Phone Rings in a Securities Fraud Case

Like the victims in The Ring, former GWG Holdings executive Bradley Heppner never suspected that pressing play could seal his fate. Charged with securities and wire fraud for allegedly misappropriating about $150 million before GWG’s bankruptcy, (which caused roughly $1 billion in investor losses), he used the consumer version of Claude to map out potential defenses and legal arguments. Federal agents later seized 31 of those AI-generated documents.

When his lawyers tried to bury the tape, Judge Rakoff refused in United States v. Heppner (S.D.N.Y. Feb. 17, 2026)**. He ruled an AI chatbot is not an attorney, sharing information with a commercial AI platform is not confidential, and forwarding the outputs to counsel later could not make them privileged.

The phone had rung. His private strategy landed in prosecutors’ hands, and he was found guilty of federal fraud counts. For anyone in the financial markets, a private brainstorm with a chatbot may be a recording waiting to be played back in court.

The Firm Chatbot Will Not Save You

Now picture a trader on a Chicago options desk facing an SEC market manipulation investigation. This time, typing late-night questions into the firm’s own AI chatbot instead of a public app. The trader’s position improves, but the curse isn’t broken.

While an enterprise tool with contractual confidentiality and no training on inputs can fix the “leaky tape” problem that sank Heppner – confidentiality, unfortunately, is only one element. Privilege also must be considered, and privilege still requires communication with a lawyer for legal advice. Courts have stressed that this relies on a trusting human relationship that cannot exist with software. So, when the trader asks AI whether order patterns look like spoofing, technically, legal counsel has not been sought.

Adding to the horror, those chats now sit on the firm’s servers as its records, which are required to be retained. Because privilege from an internal investigation typically belongs to the firm, the firm can waive it and hand the logs to SEC enforcement staff for cooperation credit. Unless counsel directs and supervises the AI use, the trader has simply traded a stranger’s tape for one the employer holds, and the SEC may be the next to press “Play”.