AI is showing up in unexpected places — from hospital rooms to church basements — as organizations across the globe find new ways to put the technology to work. At the same time, researchers and security experts are raising flags about the vulnerabilities and risks that come with these deployments.
Italian police have used an AI program called Swoads to recover stolen art and antiquities valued at over $200 million. The system helped identify more than 300 large paintings and statues along with over 165,000 smaller objects like coins, letters, and medals. One notable find was an oil painting of Saint Agenese stolen from a church near L'Aquila roughly 20 years ago, which was returned to the Diocese of Sulmona-Valva in May 2025.
In the medical field, Methodist Hospital in Omaha, Nebraska is using an AI platform called egnite CardioCare to flag patients with aortic stenosis earlier than before. The platform reads plain language from reports and categorizes severity for the structural heart team. The hospital now sees 35% more patients, and the time from first visit to treatment has dropped from eight to nine weeks down to four to six weeks.
On the security front, new research reveals that the hardware running major AI models can be hacked to trick chatbots into giving faulty or dangerous answers. Separately, a wave of AI-powered cyberattacks known as Black October exploited two critical PHP vulnerabilities, prompting the FBI and CISA to urge organizations to patch immediately. PHP powers roughly 70% of websites with a known server-side programming language.
Google's Gemini is also in the mix. DeepMind's Genie 4, built on Gemini Omni Flash, now supports audio and video prediction together, first-person navigation, and real-time camera steering — though researchers note it is still in development and can make mistakes. Meanwhile, a new research paper argues that breaking complex AI tasks into smaller pieces works better than simply making models bigger, with the approach tested across nine models ranging from a 494M-parameter local checkpoint to a frontier Opus-class model.
AI is also expanding in retail and law enforcement. About two dozen stores in the Chicago area are using Veesion's AI software on existing security cameras to analyze activity in real time and alert employees to suspicious behavior. Critics, however, raise concerns about surveillance and potential misuse. Human rights investigators are testing AI tools in places like Lebanon and Mexico to analyze large archives of violations, but families worry about how their information might be used.
The Department of War is pushing to implement AI across all levels of national defense, but experts say using reversibility alone to decide when to trust AI oversimplifies military decision-making. Risk compounds when low-consequence AI decisions feed into high-impact ones. Looking ahead, AI is expected to touch every career industry, and organizations and workers will need to adapt as the technology becomes more embedded in professional life.
Key Takeaways
- Italian police used an AI program called Swoads to recover stolen art and antiquities worth over $200 million, including 300+ large paintings and statues and 165,000+ smaller objects.
- Methodist Hospital in Omaha is using egnite CardioCare to identify aortic stenosis patients sooner, increasing patient volume by 35% and cutting treatment time from 8-9 weeks to 4-6 weeks.
- Hackers can exploit AI hardware to trick chatbots into giving dangerous or faulty answers, according to research shared exclusively with CNN.
- The Black October cyberattack wave exploited two critical PHP vulnerabilities affecting roughly 70% of websites with a known server-side programming language, prompting urgent warnings from the FBI and CISA.
- Google's Gemini powers DeepMind's Genie 4, which adds audio prediction, first-person navigation, and real-time camera steering — but remains in research with known error issues.
- A new research paper shows decomposing complex AI tasks outperforms expanding model size, tested across nine models from a 494M-parameter checkpoint to a frontier Opus-class model.
- About two dozen Chicago-area stores use Veesion's AI on security cameras for real-time shoplifting detection, though critics warn about surveillance concerns.
- Human rights investigators in Lebanon, Mexico, and other regions are testing AI to analyze large archives of violations, but families worry about data misuse.
- The Department of War is pushing AI across national defense, but experts caution that low-consequence AI decisions can compound risk when they feed into high-impact military choices.
- AI is expected to impact every career industry, requiring organizations and workers to adapt as the technology becomes more integrated into professional environments.
AI hardware can be hacked to give chatbots dangerous answers
New research shows that the hardware running major AI models can be hacked. Hackers could trick chatbots into giving faulty or dangerous answers to users. The findings were shared exclusively with CNN. This raises concerns about the safety of AI systems.
AI shoplifting detection expands in Chicago area stores
AI-powered shoplifting detection is being used in about two dozen stores across the Chicago area. The company Veesion adds AI software to existing security cameras. The system analyzes activity in real time and alerts employees about suspicious behavior. Some critics raise concerns about surveillance and potential misuse.
The promises and limits of AI in human rights work
AI is being used by transitional justice actors to help investigate human rights violations. A human rights researcher shares stories from Lebanon, Mexico and other places where AI tools are being tested. AI can help analyze large archives and find patterns, but it can also make mistakes or miss important context. Families worry about how their information might be used in AI-assisted searches.
Genie 4 adds audio and live Gemini control features
Genie 4 is the latest update in DeepMind's Genie line of world models. It can predict video and audio together from one model instead of adding sound later. The model also supports first-person navigation and real-time camera steering. Genie 4 is built on Gemini Omni Flash and is still in research. Researchers say it can still make mistakes and dialogue can have problems.
AI-powered cyberattacks fuel Black October fears
A wave of AI-powered cyberattacks called Black October has exposed serious cybersecurity threats. Hackers exploited two critical vulnerabilities in PHP known as PHP2Shell. PHP powers roughly 70% of websites with a known server-side programming language. The FBI and CISA urged organizations to patch their systems immediately. Security researchers confirmed successful exploitation in test environments, though the full impact on real-world websites remains unclear.
How can we trust AI in national defense missions?
The Department of War is pushing to implement AI across all levels of national defense. Some suggest using reversibility and consequences to determine when to trust AI versus keeping humans in the loop. However, this approach oversimplifies complex military decision-making processes. Risk compounds when low-consequence AI decisions feed into high-impact decisions. For AI to be trusted, it must account for data, context, model, and actions behind every answer.
Decompose, Don't Expand: Capability-Independent AI Systems by Construction
A new research paper argues that decomposing complex AI tasks is better than expanding model size. The study introduces parity by construction, where model size no longer affects performance on extracted sub-tasks. The approach was measured across nine models spanning a 494M-parameter local checkpoint to a frontier Opus-class model. The method works across physics setup, motion retargeting, object-model editing, and Web-CLI tools. The paper provides a way to locate the cut between deterministic and generative components.
Italian Police Used AI to Identify $200 M. Worth of Stolen Art and Antiquities
Italian police used an AI program called Swoads to identify stolen art and antiquities worth over $200 million. The program recovered more than 300 large paintings and statues, and over 165,000 small objects like coins, letters, and medals. In fall 2024, Swoads identified an oil painting stolen from a church near L'Aquila 20 years prior. The painting of Saint Agenese was returned to the Diocese of Sulmona-Valva in May 2025.
Looking at the future of artificial intelligence in the workplace
AI is expected to impact every career industry to some degree. The article explores how artificial intelligence will shape the future of work across various fields. Organizations and workers will need to adapt as AI becomes more integrated into professional environments. The extent of impact will vary by industry and role.
Methodist uses AI to flag severe heart valve cases sooner
Methodist Hospital in Omaha, Nebraska is using an AI platform called egnite CardioCare to identify patients with aortic stenosis sooner. The platform reads plain language from reports and categorizes patients by severity for the structural heart team. A man in his 60s was treated with a surgical aortic valve before he would have been seen in the usual clinic. The program has allowed Methodist to see 35% more patients and reduced the time from first visit to treatment from eight to nine weeks down to four to six weeks.
Sources
- AI hardware can be hacked so chatbots provide dangerous answers, research shows
- AI-powered shoplifting detection expands in Chicago-area stores, but some raise concerns
- The room she kept ready, and the promises of AI
- Genie 4 adds audio and live Gemini control
- AI-powered cyberattacks fuel 'Black October' fears
- How can we trust AI in national defense missions?
- Decompose, Don't Expand: Capability-Independent AI Systems by Construction
- Italian Police Used AI to Identify $200 M. Worth of Stolen Art and Antiquities
- Looking at the future of artificial intelligence in the workplace
- Methodist uses AI to flag severe heart valve cases sooner
Comments
Please log in to post a comment.