Fleets own what their AI dash cams miss, defense attorney warns

Days after a Dallas County judge upheld a $604 million jury verdict against C.H. Robinson, a trucking defense attorney told fleets in Charlotte that their AI dash cams have opened a new door for plaintiffs: risky driving the camera recorded but the AI never flagged.

Doug Marcello, a shareholder at Saxton & Stump, said nuclear verdicts rarely grow out of the crash facts themselves. Plaintiff attorneys hunt for evidence of systemic failure, and AI-only camera systems hand them a fresh place to look — footage the fleet's own safety team never reviewed or coached.

Marcello made the case at the American Trucking Associations' Management Conference & Exhibition in Charlotte, North Carolina, on a panel with Sean Ritchie, a vice president at Solera Fleet Solutions, which sells video safety systems with a human review layer. The setting matters: motor carriers ranked lawsuit abuse reform as their top concern in the American Transportation Research Institute's 2026 Top Industry Issues report, released Oct. 18 at the conference, and it ranked second overall.

What does the fleet actually own?

Everything on the video, per Marcello. "If you have the activity of the driver captured by the video, you're going to be responsible for it," he said. "And the defense or being able to say, wait, well, AI didn't pick it up is not going to be the response. It's not going to be an excuse."

How AI changed camera review

The mechanics are simple. Early video systems recorded an event when an accelerometer sensed G-forces past a set threshold, Ritchie said, and reviewers went through each clip frame by frame. As recently as five or six years ago, almost every video passed through a human layer before it reached the fleet.

Today's cameras use computer vision trained to label risky behavior, and many send videos straight into the coaching workflow with no human review. Federal rules cover hardware placement only: FMCSA's windshield rule allows safety tech mounted no more than 8.5 inches below the top of the wiper-swept area. Nothing covers how video is reviewed.

With no government standard, Marcello said, juries will set one. "It will be an economic governance by what you do or fail to do with that AI," he said.

The false negatives fleets don't see

Ritchie opened the session with a show of hands. Who uses dash cams? Every hand went up. Who knows their false-positive rate? A few dropped. Who knows their false-negative rate — risky behavior on video the AI didn't label? Every hand went down.

Solera analyzed AI-triggered videos and found up to 30% held other risk the AI didn't flag, Ritchie said. The company hasn't published the sample size or method, and has cited a version of the figure since at least August 2025. Rates vary by vendor and risk type; drowsy driving is the most common miss. Ritchie tied the misses to vendors tuning out false alerts: because the industry complained false positives were too high, vendors "started to dial down the sensitivity," creating more opportunity for real risk to go unlabeled.

Why unflagged video matters in court

The systemic-failure argument grows out of the reptile theory, which Marcello described as "primarily a strategy for maximizing verdicts based upon engendering fear in the jury."

"If we have a situation where there is risky behavior not caught or not flagged on the video by AI that goes uncorrected, even if that doesn't relate to what caused the accident... it will be skewed in terms of being indicative of the lack of a safety culture and thus a systemic failure," he said.

A mock trial with a Philadelphia focus group showed how jurors judge follow-through. Jurors heard about a 35-truck carrier whose driver hit a line of traffic and killed two people, with several overspeeds and hard brakes the prior week that a lone safety director hadn't caught. "That mock jury was totally unsympathetic," Marcello said. "And the gist of it was, if you're on the road, you better be paying attention." The same logic applies to documented unsafe driving on fleet video, "whether they're picked up by AI or not."

Retention policies keep the evidence on hand. Providers typically hold no-risk video for 30 days at most and flagged video for six months or longer — so a clip flagged for a seat belt violation can sit in the system carrying unflagged drowsy driving.

Some companies avoid analyzing data for fear of discovery. Marcello calls this the discovery fallacy: "The fallacy is we can prevent discovery by not looking at it. It'll be discovered, the data itself." Not looking only means the plaintiff finds it first.

Three ways vendors handle review

Ritchie split vendors into three camps: those piping camera video straight into fleet workflow, those using humans only to strip false positives, and those screening for both false positives and false negatives before video reaches the fleet.

Caveats deserve note. The argument comes from a vendor selling human review, and neither panelist cited a court ruling that turned on an AI-missed event. And Ritchie said dropping AI wouldn't help either — human review alone, "the old way of doing it," is no better, "just as the AI alone is not better."

"Even though I sell AI cameras, that's literally my job," Ritchie said. "I'm seeing a situation where we moved really fast, got a bunch of AI only cameras throughout the entire industry, and not one time have I heard anybody talk about a false negative. That can't be the case anymore."

The C.H. Robinson case shows the trajectory. The jury assigned 23% of fault to the broker and 45% to the driver, Gorgonio Gonzalez, finding him "subject to control by C.H. Robinson as to the details of the mission." Marcello, who pulled the docket filings, said plaintiffs argue that makes Gonzalez a borrowed employee — combining the shares above 50% and leaving the broker owing the full verdict. Cases like it, he said, will define the standard of care one verdict at a time.