Investigators face new struggles in the age of AI and technology

AI investigation challenges surge as faked evidence baffles U.S. detectives

Sarah Mills | GlobalBeat

Local detectives say fabricated videos and cloned voices now routinely derail criminal cases across the United States.

Springfield Police Commissioner Cheryl Claprood told reporters her department had dropped 3 probes this year after AI-generated alibis surfaced.

“This is the fastest-growing obstacle we have ever confronted,” she said at a briefing on Wednesday. “Evidence can be invented in minutes.”

Detectives nationwide report similar disruption. The FBI counted 2,700 AI-facilitated fraud attempts in 2023, up from fewer than 200 in 2021.

“We are losing the ability to trust audio, images and documents,” said Hany Farid, a digital forensics scholar at the University of California, Berkeley. “That undermines every stage of an inquiry.”

Criminals now download voice-cloning apps, call victims with cloned pleas for cash and feed investigators deepfake videos that place suspects miles from crime scenes, officials said.

A Massachusetts state task force said one synthetic clip had delayed a human-trafficking indictment for six months while analysts tried to prove it was false.

The result is longer investigations, higher lab costs and growing jury scepticism, according to prosecutors.

State attorneys met last month in Denver to draft model legislation that would require disclosure of AI-generated material introduced in court.

Defence lawyers counter that such rules risk stifling legitimate exculpatory evidence.

“AI can expose police misconduct,” said Jeffrey Harris, a public defender in Boston. “Blanket distrust hurts innocent people.”

The disarray comes as federal funding for forensic research has fallen 12 percent since 2020, department figures show.

Private labs now charge up to $5,000 per file to detect synthetic media, costs many counties cannot afford.

Detective Bureau Chief Jorge Fontanez of Hampden County said his 28 investigators share one ageing workstation equipped with deepfake detection software. “We are outgunned,” he said.

Experts trace the problem to free online tools that create convincing fakes with a single photograph or a 30-second voice sample.

Open-source code released by startup Stability AI last year slashed the time needed to produce a synthetic face from hours to 90 seconds, researchers at Stanford said.

Cyber-criminals advertise deepfake services on Telegram, promising “untraceable alibi videos” for $150, according to screenshots obtained by the Federal Trade Commission.

The rush has forced law-enforcement labs to revise training manuals they finished updating only three years ago.

“We issue new guidance and it is obsolete within weeks,” said Matthew Robinson, who oversees digital evidence standards at the National Institute of Standards and Technology.

He said suppliers of detection software typically update algorithms monthly to keep pace, creating compatibility problems with police computer systems that upgrade twice a year.

Police chiefs told Congress in March that delays caused by authentication disputes can embolden criminal networks and erode public trust.

Senator Richard Blumenthal, a Connecticut Democrat, said he would re-introduce a bill that sets a 48-hour deadline for federal labs to verify or refute synthetic media.

The measure also proposes $30 million in annual grants for local agencies to buy detection tools.

Republicans on the Senate Judiciary Committee oppose new spending, arguing that tech companies that released the software should fund the response.

“Platforms created this mess,” ranking member Chuck Grassley said at a hearing. “They can pay to clean it up.”

Meta, Google and Amazon declined to comment on that claim when contacted on Thursday.

Privately, industry lobbyists said detection funding should come from government because police ultimately choose which evidence to rely on.

Academics warned the controversy diverts attention from simpler policing reforms that could blunt AI abuses.

“Check the origin of every file, insist on chain-of-custody logs and corroborate alibis with witnesses,” Berkeley’s Farid said. “Good detective work has not changed.”

Still, officers on the ground said the volume of digital material makes manual checks impractical.

A single homicide in Springfield last March produced 1,800 video clips from doorbell cameras, patrol cars and social media, detectives said.

Lab technicians had to screen each file for deepfake artefacts while processing DNA and ballistics, stretching the inquiry to four months.

Prosecutors ultimately secured a plea deal, but the victim’s family complained the wait prolonged their grief.

“We do not have the staff to repeat this in every case,” Commissioner Claprood said.

Background

Police have confronted fake evidence for decades, from doctored photographs to planted documents, but the process once required technical skill and expensive equipment.

Digital editing in the 1990s allowed amateurs to superimpose faces onto incriminating scenes, yet analysts could still spot inconsistencies in lighting or metadata.

Widespread generative AI that emerged in 2022 removed those hurdles, enabling anyone with a smartphone to create footage that passes casual inspection.

Unlike earlier software, modern algorithms learn from millions of images and audio samples, reproducing skin texture, lip movement and vocal cadence down to micro-expressions.

The Department of Homeland Security warned in 2021 that hostile states might deploy deepfakes to discredit U.S. elections, but criminal adoption has moved faster than foreign interference.

What’s Next

The Senate Judiciary Committee plans a vote on the deepfake detection bill in September, while states including California and Texas weigh parallel measures that could set conflicting standards.

Police leaders said they expect at least 15 departments to pilot free federal detection tools by December, with the first results due early next year.

If adoption lags, prosecutors predict more mistrials, generous plea bargains and cold cases as courts wrestle with evidence they can neither trust nor ignore.