The measures signed to the law this summer target the use of deepfake images or voices in Pennsylvania's financial fraud, highlighting the ongoing struggle in the state's court system to deal with the deluge of generative artificial intelligence that has recently hit the internet.
As the internet becomes increasingly flooded with AI, authorities have been working to counter more insidious uses, from useful information outlines on Google's search to images and videos that are fully manufactured on social media.
Last year, Gov. Josh Shapiro signed a law banning artificially generated images and videos depicting child sexual abuse. The state's first arrest under that law took place in York County in April, according to the Attorney General's Office.
Then, on July 7, Shapiro signed another bill, criminalizing using AI to create non-consensual “Deepfake Images, videos, audio clones, etc.) to fraud or harm Pennsylvanians.
Luzerne County officials say they have other concerns about how AI could be misused, but generally support the measures.
Vito Deluca, Luzerne County's first assistant lawyer and chief artificial intelligence officer, said the changes were positive and would help address the real issue.
“I have no problems with that kind of law at all. I think it's going to be a problem when I try to regulate some of the other areas,” Deluca says. “There are many government agencies considering predictive policing. They predict where there is the most likely chance that crime will occur. I call a lot of that snake oil because it's very difficult to predict the future.”
Meanwhile, district attorney Sam Sangedors is “optimistic” about changes in the law, but he still says he is concerned, as he has continued difficulties in chasing criminals operating overseas, even with the help of national organisations such as the FBI.
“My hesitation is that when the subject of the law distributes child pornography and fraud from India, China or Russia, no matter what law you pass, there's still nothing we can do about it,” SangueDolce said. “That's really my big concern because I see older people being scamed for thousands of dollars every day.”
Raise with AI
The rapid escalation in the use of AI has presented many challenges to the police and the courts.
The rise of so-called virtual assistant applications such as Openai's ChatGpt and Google Gemini gave lawyers an automated means of handling a variety of routine tasks, from compiling electronic discovery materials to draft legal briefs and conducting legal investigations.
However, the tendency of AI to “hastised” presents information manufactured as fact – a continuing concern. Last year, US District Judge Karoline Mehalchick cited these concerns when he issued an order instructing lawyers to disclose specific AI tools used in preparing documents and identify which segments of the document were programmatically created.
New Act on Deepfakes – This was co-hosted by Senator Rosemary M. Brown, R-40 and Middle Smithfield Twp. – Makes a three-fold felony of committing fraud or causing injury using fake content generated by AI. For example, use human voice breeding to convince them to send money.
“In Pennsylvania, we are leading AI and take advantage of the economic and technical benefits that come with it, but we also take a thoughtful and proactive approach to protecting Pennsylvanians,” Shapiro said in a statement when signing the law. “By signing this bill to the law, I send you a clear message that if you use AI to fraud or misuse Pennsylvanians, you will be accountable. My administration is committed to cracking down on fraudsters, protecting consumers, ensuring Pennsylvanians of all ages feel safe and confident in the digital age.”
In their announcement, state officials said that elders' financial exploitation is a continuing increase in Pennsylvania, and that the Department of Aging received nearly 18,500 reports in the 2023-3024 fiscal year.
“With the rise in financial crimes, including artificial intelligence, it's becoming more difficult to tell what's real and what's fake,” Bank and Securities Director Wendy Spicher said in a statement. “It's important that consumers remain vigilant and notified if they have questions or concerns, knowing that help is just a call or click.”
AI supports law enforcement
Criminals aren't just using emergency technology.
Wilkes Barre TWP throughout the summer. Police have released two photos of the suspect who wanted to try to cash out a $6,000 check at the M&T Bank branch.
One of the photos was a relatively grainy surveillance image taken directly from a bank camera. The other was the same photo as “enhanced using AI,” the department wrote on Facebook.
Some users have pointed out in entertainment that the enhanced image appeared to portray actor Brad Pitt, or even described as “a Dollar Tree version of Leonardo DiCaprio.”
Wilkes Barre TWP. Police patrolman Matt Godlusky, a member of the department's technical services unit, likened the use of AI photo enhancements to police using computers to reconstruct the faces of skeletonized bodies.
“It's a tool, it's not necessarily something that depends on,” he said. “When using it to seek suspects in a probably blurry photo, we're not saying that the image is definitely a person. But more or less, it acts as a recognition tool.”
Godlewski pointed out that many of the surveillance images obtained by police are thin and pixelated, making them difficult to recognize. Social media users are known to criticize image quality, and removing images from the space can be more clear than faces taken inside a local store.
“Unfortunately, that's true,” Godlesky said. “Despite our demands that they get a better camera, they're tied to what they have inside.”
As a result, Detective Robert Caperell created an enhanced image with what the department is considered to be the first use of AI, he said.
“We could use image enhancements when necessary to help us develop these leads better,” says Godlewski. “If you have a photo and it's a pixelated face, using AI may help you better develop a recognizable face.
The dangerous aspects of AI
Deluca, the county's chief AI officer, said generative AI has become a “overnight feeling” because its images and words are flooded with social media, but the other unattractive aspect is where he is real concern.
“Insidious and dangerous artificial intelligence is an AI that is not chatgpt,” Delca said.
The technology can be used in applications ranging from facial recognition and recidivism prevention to attempts to predict where a crime will occur before it occurs, he said.
He said that problems can arise when AI uses inappropriate standards that have a “different impact” on classes of protected people.
“The problem is that when the model is built on biased data, it produces biased results,” Deluca said. “It could lead to visits from the Department of Justice and lawsuits from the ACLU without first guiding your head and adopting an artificial intelligence solution and actually researching what the impact is.”
Citing Maxim in “Garbage, Garbage Output,” Delka said that AI programs programmed with data showing high crime areas predict more crime in the same place.
“It's not rocket science,” Delca said. “If you have artificial intelligence, make recommendations or decisions that humans should make, make sure they're correct.”
In a December 2024 report entitled “Artificial Intelligence and Criminal Justice,” the Department of Justice advised solid governance to ensure that AI is implemented fairly in courts.
“The use of AI in the criminal justice system must be governed by robust human-centered policies and procedures that explain the permitted and prohibited use of AI systems, the data used to train the system, the accuracy of output, the accuracy of monitoring, the frequency of monitoring, and the risks of using a particular AI system,” the report states.
Evolving technology
The surge in content generated by AI also raised the prospect that images and videos obtained as evidence could be altered or manipulated during criminal investigations.
Deluca noted that although AI content often conveys its fraud, it could potentially be portrayed by six fingers, so technology is constantly improving and producing more persuasive images.
“As technology improves, it will be much more difficult if you are trying to identify AI models,” Deluca said.
SangueDolce said the issue of AI-based evidence that entering criminal cases is concerning mobile phone video compared to surveillance footage, as police obtain videos from ring cameras and other security systems early in the investigation.
“That's a concern, but from a research perspective, getting the video early is not too worrying, as it's difficult to inject AI into that source,” says SangueDolce. “But the cell phone videos that come out, that's where the problem really lies.”
Prosecutors pointed out that cell phone videos that are fragmented or provide limited angles can give them an incomplete picture of what actually happened, as in the case of suspected police brutality.
“They play 15-second cell phone video on the news. Then, if you get the other five seconds, it's a completely different story,” SangueDolce says.
He hopes applications will be released soon to assist in the detection of content manipulated by artificial intelligence, and that police and prosecutors must focus on certifying their evidence.
“From my perspective, it's really whether you got the video from a reliable source,” SangueDolce says. “That's the investigation part. And as for the trial part, what is it that's going to be injected into the trial to confuse the courts and ju judges? Is it real?”
Initially published:
