Law360 (November 13, 2025, 3:13 PM EST) -- Veronica Finkelstein This article is part of a monthly column analyzing what modern movies can teach attorneys about the practice of criminal law. This installment explores how the film "Roofman" illuminates the evidentiary doctrine of modus operandi — and how emerging technologies like artificial intelligence complicate its use in court.

In the crime dramedy "Roofman," released last month, Channing Tatum portrays Jeffrey Manchester, a former U.S. Army sergeant who became infamous for a string of rooftop burglaries across nine states.

As unlikely as it sounds, this film is based on real events. Between 1998 and 2000, the real-life Manchester robbed dozens of fast-food restaurants — primarily McDonald's — by drilling through their roofs, hiding in restrooms until employees arrived and then locking the employees in walk-in freezers while he emptied the registers. His polite demeanor and consistent pattern of behavior earned him the nickname "Roofman."

The film dramatizes Manchester's signature style of burglary, raising a central legal question: When does a pattern of behavior become so distinctive that it can be used to prove identity in a criminal trial?

Under Rule 404(b) of the Federal Rules of Evidence, prior bad acts are generally inadmissible to show propensity.[1] These same acts, however, may be admitted to prove identity if the acts share a signature or unique quality.[2] This signature quality is what courts refer to as modus operandi.[3] To qualify as modus operandi, the pattern must be sufficiently unusual and distinctive so as to be like a fingerprint.[4]

The definition of what constitutes modus operandi was tested in U.S. v. Edwards, where the defendant was charged with multiple bank robberies.[5] The government introduced evidence of prior robberies to establish identity under Rule 404(b), arguing that the defendant's method of committing the crimes was so distinctive it functioned as a signature.[6]

Indeed, the robberies shared several specific features: Each robbery occurred around 5:30 p.m. in northeastern Illinois during a two-month period. In each robbery, there were two to three perpetrators wearing a hat or hood to cover their face, and gloves or shirtsleeves to cover their hands.[7]

Before initiating the robbery, one individual engaged in conversation with a store employee.[8] The other individuals waited until all customers had exited the premises, at which point a perpetrator brandished a firearm, loaded it and aimed it at an employee as another accomplice secured the front entrance.[9] The staff were then escorted to the rear of the store, where one was compelled to unlock the safe.[10]

The perpetrators filled black plastic bags with cellphones; attempted to eliminate traces of their presence, sometimes by requesting surveillance recordings; and fled through a rear exit to reach their getaway vehicle.[11]

Antonio Edwards was charged with a variety of offenses arising from three robberies.[12] He pled guilty to a majority of the charges relating to two of the robberies, but went to trial on charges for the third.[13] At trial in the U.S. District Court for the Northern District of Illinois, the government introduced evidence from all three robberies to prove identity, arguing that the robberies shared a signature character.[14] The defense objected, arguing that the similarities were not distinctive enough and that the evidence was prejudicial.[15]

On appeal, the U.S. Court of Appeals for the Seventh Circuit in 2022 agreed with the government. It found that the characteristics of the three robberies, considered in total, suggested a sufficiently unique modus operandi.[16] The court noted that although certain elements of the robberies might have appeared typical, their combined presence revealed a unique operational pattern specific to the defendant and his accomplices.[17]

Although individual traits, such as location, use of a getaway car or takeover tactics, may not independently prove a modus operandi, their collective recurrence can establish a recognizable pattern.[18]

This case remains a touchstone for prosecutors seeking to admit prior bad acts to prove identity. But it also illustrates the limits of modus operandi pattern evidence: Individual similarities are not enough. Similarities must be aggregated to create a truly signature pattern.

The film "Roofman" underscores this principle. Manchester's rooftop entry, freezer confinement and apologetic tone were not just similar — viewed in totality, they were unique. Human investigators were able to identify and show a pattern between the various robberies.

Manchester was caught because his modus operandi of entering through the roof was recognized and used by law enforcement to track him down. Manchester attempted to rob a McDonald's restaurant. As he entered the building through the roof, it triggered the alarm. Manchester fled without taking any money from the store. The police suspected that Manchester would strike again at another McDonald's in the area. They staked out the area and caught him doing exactly that — following the distinctive pattern law enforcement had predicted he would follow.

At the time Manchester was first tried in 2000, artificial intelligence was in its nascent stages of development.[19] Now, generative AI tools are becoming commonplace in the legal industry.[20] In the future, AI might be used to develop modus operandi evidence through a practice known as crime series detection.

Crime series detection refers to the analytical process of identifying patterns among criminal incidents that suggest they were committed by the same individual or group.[21] This task involves examining similarities in the specific methods or behaviors used during the commission of crimes.[22] For example, a series of burglaries might share traits such as occurring on weekday afternoons, targeting homes with unlocked back doors and involving minimal property damage.

Recognizing these patterns helps law enforcement agencies link cases that might otherwise appear unrelated, potentially narrowing down suspects and guiding investigative resources more effectively.

AI has emerged as a tool in enhancing crime series detection.[23] Traditional methods rely heavily on manual database queries and human intuition, which can be time-consuming and prone to oversights. Some AI algorithms can simultaneously identify both the relevant variables that define a modus operandi and the clusters of crimes that share those traits.[24]

This dual capability allows AI systems to uncover hidden connections across large datasets, revealing crime series that might elude human analysts.[25] Cities like New York have begun integrating such technologies into their investigative workflows, using them to generate leads and prioritize cases.[26]

Despite its promise, the use of AI in crime series detection raises important questions about its role in the courtroom.[27] Although these algorithms can suggest that certain crimes are connected, they do not establish individual culpability. The patterns identified by AI are probabilistic and inferential, not definitive proof of guilt.[28]

Therefore, while AI-generated crime series detection may inform investigations, the patterns suggested by AI tools must be corroborated with traditional evidence — such as eyewitness testimony, forensic data or surveillance footage — before being introduced at trial. The legal system demands a higher standard of proof than statistical similarity, and courts have yet to address the admissibility of modus operandi evidence gathered exclusively through the use of AI tools.

Nevertheless, AI's potential to support the development of modus operandi evidence is significant. By systematically analyzing behavioral patterns, AI can help prosecutors build narratives that contextualize a suspect's actions within a broader series of crimes.

As legal standards evolve, and AI tools become more transparent and interpretable, we may see a future where crime series detection plays a more prominent role in courtroom proceedings — provided it is used responsibly and in conjunction with other forms of evidence.

As "Roofman" reminds us, modus operandi evidence has long offered a powerful way to connect a defendant to a crime, especially when the conduct is so distinctive it seems to speak for itself. But the evidentiary landscape is shifting. Manchester's rooftop entries and freezer confinements were pieced together into a distinctive pattern by human investigators. In the future, this work might be done by AI.

Veronica J. Finkelstein is an associate professor at Wilmington University School of Law. She previously served as an assistant U.S. attorney in the U.S. Attorney's Office for the Eastern District of Pennsylvania.

The opinions expressed are those of the author(s) and do not necessarily reflect the views of their employer, its clients, or Portfolio Media Inc., or any of its or their respective affiliates. This article is for general information purposes and is not intended to be and should not be taken as legal advice.

[1] Fed. R. Evid. 404(b).

[2] See U.S. v. Miller , 589 F.2d 1117 (1st Cir. 1978); Oliphant v. U.S. , 525 F.2d 505 (9th Cir. 1975).

[3] See Hurst v. State , 929 A.2d 157 (Md. 2007).

[4] See State v. Lowe , 634 N.E.2d 616, 619-20 (Ohio 1994) ("A certain modus operandi is admissible not because it labels a defendant as a criminal, but because it provides a behavioral fingerprint which, when compared to the behavioral fingerprints associated with the crime in question ... ").

[5] United States v. Edwards , 26 F.4th 449, 451-52 (7th Cir. 2022).

[6] Id. at 455.

[7] Id.

[8] Id.

[9] Id.

[10] Id.

[11] Id.

[12] Id. at 452.

[13] Id. at 452-53.

[14] Id. at 453.

[15] Id.

[16] Id. at 455.

[17] Id.

[18] Id.

[19] John M. Glionna, "Roofman" Gets the Blame for 38 Robberies in 9 States, Los Angeles Times (April 26, 2000).

[20] Xavier Rodriguez, Symposium: Artificial Intelligence: Judicial Regulation on the Use of AI, 109 The Advocate 17, 17 (2024).

[21] Brandon L. Garrett & Cynthia Rudin, The Right to a Glass Box: Rethinking the Use of Artificial Intelligence in Criminal Justice, 109 Cornell L. Rev. 561, 584 (2024).

[22] Id.

[23] Paula Helm & Thilo Hagendorff, Black Box Artificial Intelligence and the Rule of Law: Beyond the Prediction Paradigm: Challenges for AI in the Struggle Against Organized Crime, 84 Law & Contemp. Prob. 1, 4 (2021).

[24] Garrett & Rudin, supra note 20.

[25] See Iria Giuffrida, Symposium: Rise of the Machines: Artificial Intelligence, Robotics, and the Reprogramming of Law: Liability for AI Decision-Making: Some Legal and Ethical Considerations, 88 Fordham L. Rev. 439, 441 (2019) (noting that AI detection of patterns is so attuned that it may even be used to predict future behavior); see also Chritopher Rigano, Using Artificial Intelligence to Address Criminal Justice Needs, 280 Nat'l Inst. Just. J. 37, 39 (2018).

[26] Garrett & Rudin, supra note 20.

[27] See Sherman J. Clark, Confronting Algorithms: Conscience Catching in the Criminal Trial and Beyond, 57 U. Mich. J.L. Reform 787, 787-789 (2024).

[28] See Chris Schwegmann, The Future of AI in Law Embracing the Hallucinations, 51 Litigation 42, 43 (2025).

For a reprint of this article, please contact reprints@law360.com.