Google Health's diabetic retinopathy AI faced real-world issues in Thailand clinics
Google Health deployed a deep-learning system to screen for diabetic retinopathy in 11 clinics across Thailand. The AI was highly accurate in lab tests but rejected over a fifth of eye scans in real-world conditions due to poor lighting and slow internet. Patients whose scans were rejected had to visit specialists at other clinics, causing inconvenience and frustration for nurses. Google Health is now working with local staff to improve the system's workflow.
- Company involved
- Google Health
10 source articles · read the reporting →
Deep-Live-Cam enables real-time face swapping, raising fraud concerns
A software package called Deep-Live-Cam has gone viral for enabling real-time face swapping on webcams using a single photo. The tool, which is free and open source, allows anyone to impersonate another person's appearance in a video chat, raising concerns about remote deception and fraud. The article notes that similar technology has been used in a $25 million heist in Hong Kong and in audio deepfake scams.
- AI system involved
- Deep-Live-Cam
7 source articles · read the reporting →
Stable Diffusion amplifies racial and gender stereotypes in generated images
An analysis by Bloomberg of over 5,000 images generated by Stability AI's Stable Diffusion found that the text-to-image model amplifies racial and gender stereotypes. The model overrepresented lighter-skinned men in high-paying jobs and darker-skinned people in low-paying jobs, and underrepresented women in positions of power. Stability AI acknowledged the inherent biases in its models and stated it is working on mitigation.
- Company involved
- Stability AI
- AI system involved
- Stable Diffusion
8 source articles · read the reporting →
OpenAI's CLIP vision system fooled by handwritten notes
OpenAI researchers discovered that their CLIP computer vision system can be deceived by handwritten labels placed on objects. The system's multimodal neurons respond to text as well as images, causing it to misidentify objects. The attack, called a typographic attack, is a research finding and not a deployed system. No actual harm occurred.
- Company involved
- OpenAI
- AI system involved
- CLIP
10 source articles · read the reporting →
Apple's Enhanced Visual Search raises privacy concerns over default data sharing
Apple's iOS 18 update introduced an 'Enhanced Visual Search' feature that automatically shares encrypted photo data with Apple to identify landmarks. The feature is enabled by default, requiring users to manually opt out, which has sparked privacy concerns. Critics argue it should be opt-in, given Apple's usual privacy standards. The article reports on the feature's design and the resulting debate, not on any specific harm to an individual.
- Company involved
- Apple
- AI system involved
- Enhanced Visual Search
5 source articles · read the reporting →
WebinarTV secretly records Zoom calls and turns them into AI podcasts
WebinarTV, a company that bills itself as a search engine for webinars, is secretly scanning the internet for Zoom meeting links, recording the calls, and turning them into AI-generated podcasts for profit. People only found out their calls were recorded when WebinarTV contacted them to promote its services. The recordings may put call participants at risk.
- Company involved
- WebinarTV
4 source articles · read the reporting →
DeepSeek exposed user data via open ClickHouse database
Cloud security firm Wiz discovered a ClickHouse database belonging to DeepSeek that was open to the internet without authentication, containing over a million lines of logs with chat histories, secret keys and backend details. Wiz disclosed the breach to DeepSeek, which promptly locked down the database. The incident highlights security risks in rapidly deploying AI services.
- Company involved
- DeepSeek
- AI system involved
- DeepSeek-R1
5 source articles · read the reporting →
Researchers jailbreak Stable Diffusion and DALL-E 2 to generate disturbing images
Researchers from Johns Hopkins and Duke universities developed a method called SneakyPrompt that uses reinforcement learning to bypass safety filters in text-to-image AI models. The technique allowed them to generate images of nudity and violence from Stable Diffusion and DALL-E 2. OpenAI has since fixed the vulnerability in DALL-E 2, but Stable Diffusion 1.4 remains vulnerable. Stability AI says it is working with the researchers to improve defenses.
- AI system involved
- Stable Diffusion 1.4 and DALL-E 2
5 source articles · read the reporting →
US CBP deploys CBP One app using facial recognition for asylum seekers amid privacy concerns
The article reports that U.S. Customs and Border Protection quietly deployed the CBP One mobile app at the Mexico border. The app uses facial recognition and geolocation to collect and verify information on asylum seekers before they enter the United States. Privacy experts warn that the app poses risks of persistent surveillance and that the facial recognition algorithm is unreliable for people of colour. A previous CBP facial recognition pilot was hacked, exposing images. CBP says the app is voluntary and data is secure.
- Company involved
- U.S. Customs and Border Protection
- AI system involved
- CBP One
8 source articles · read the reporting →
N-Tech.lab's FindFace used to identify St Petersburg metro passengers without consent
Egor Tsvetkov photographed passengers on the St Petersburg metro without their permission and used N-Tech.lab's facial recognition service FindFace to match their faces to public Vkontakte profiles. He published the results in an art project called 'Your Face is Big Data', saying he wanted to show how 'digital narcissism' can lead to stalking. Privacy advocates said the project was ethically problematic because the subjects had not consented and their identities were exposed. FindFace had been launched by N-Tech.lab in February 2016.
- Company involved
- N-Tech.lab
- AI system involved
- FindFace
8 source articles · read the reporting →
AI image generators produce misleading election images, study finds
A study by the Center for Countering Digital Hate found that leading AI image generators, including Midjourney, DreamStudio, ChatGPT Plus, and Microsoft Image Creator, could be manipulated to create misleading election-related images. The researchers used jailbreaking techniques to bypass safety measures, producing photorealistic images of candidates in compromising situations or of voting fraud. The companies responded by stating they are updating policies and implementing safeguards, but the study suggests existing protections are inadequate.
- Company involved
- Midjourney, Stability AI, OpenAI, Microsoft
- AI system involved
- Midjourney, DreamStudio, ChatGPT Plus, Microsoft Image Creator
8 source articles · read the reporting →
AI deepfakes disrupt Bangladesh's election
Affordable deepfake tools for $24 a month are being used to generate deceptive videos targeting voters in Bangladesh's election. The technology enables the creation of realistic fake content that could mislead the electorate. The full extent of the impact and response from authorities is not yet known.
8 source articles · read the reporting →
Seoul Metropolitan Government rolls out AI CCTV cameras to prevent suicides
The Seoul Metropolitan Government has rolled out AI-enabled CCTV cameras on Han River bridges to identify people at risk of suicide. The system uses deep learning to analyse behavioural patterns and alerts rescue teams. Experts and Privacy International have raised concerns that it is invasive and could be misused because it captures biometric data without explicit consent. The article also considers whether such technology could be implemented in India.
- Company involved
- Seoul Metropolitan Government
8 source articles · read the reporting →
Google AI Overviews generate erroneous search summaries
In May 2024, Google launched AI Overviews, a feature in Search that generates AI-powered summaries. Shortly after, users reported odd and erroneous overviews for some queries, including satirical or nonsense results. Google acknowledged the issues in a blog post and stated they made more than a dozen technical improvements to reduce inaccuracies. The company said that less than one in 7 million queries resulted in a content policy violation.
- Company involved
- Google
- AI system involved
- AI Overviews
10 source articles · read the reporting →
Baltimore schools monitor student laptops for suicide signs using GoGuardian Beacon
Baltimore City Public Schools uses GoGuardian Beacon software to monitor student laptops for signs of suicide. Since March 2021, the system has flagged 786 alerts, with nine students taken to emergency rooms. Privacy advocates warn the monitoring could lead to disciplinary actions, outing of LGBTQ students, and disproportionately affect disadvantaged students. School officials defend the practice as a safeguard.
- Company involved
- Baltimore City Public Schools
- AI system involved
- GoGuardian Beacon
10 source articles · read the reporting →
Network Rail denies using AI cameras to detect passengers' emotions
Network Rail conducted a trial of AI cameras at major stations in 2022 that analysed demographic details and reportedly assessed emotions. Documents obtained by Big Brother Watch indicate the system was capable of determining whether a passenger was happy, sad or angry, with potential use for measuring satisfaction and advertising. Network Rail denied that any emotion analysis took place and stated the image analysis for demographic details has ended. Big Brother Watch has submitted a complaint to the Information Commissioner about the trial.
- Company involved
- Network Rail
- AI system involved
- Amazon Rekognition
7 source articles · read the reporting →
Oxford Town Centre CCTV dataset used without consent for AI research
The Oxford Town Centre dataset is a CCTV video of pedestrians in Oxford, England, captured from a public surveillance camera without the knowledge or consent of the approximately 2,200 people shown. The footage was used in over 60 research projects, including commercial research by Amazon, Disney, and Huawei, for developing facial recognition, sex classification, and social distancing algorithms. The dataset was taken down in June 2020, but no remediation was provided to the individuals depicted.
- Company involved
- University of Oxford
- AI system involved
- Oxford Town Centre dataset
5 source articles · read the reporting →
Big Tech companies used YouTube videos to train AI without consent
Proof News found that subtitles from 173,536 YouTube videos were used by companies including Anthropic, Nvidia, Apple, and Salesforce to train AI models. The dataset, called YouTube Subtitles, was created by EleutherAI and published in 2020. Creators were not aware and some have expressed frustration, calling it theft. The companies have acknowledged using the dataset but argue it was publicly available.
- Company involved
- Anthropic, Nvidia, Apple, Salesforce, Bloomberg, Databricks
- AI system involved
- Claude, OpenELM
10 source articles · read the reporting →
French police secretly used Briefcam facial recognition since 2015
The French National Police has been using Briefcam's Video Synopsis software since 2015, according to internal documents obtained by Disclose. The software, which includes facial recognition capabilities, was deployed without the required data protection impact assessment or notification to the CNIL. The Ministry of the Interior concealed the use of this tool for eight years. The police hierarchy allegedly used the software for facial recognition without judicial requisition, and the DGPN did not respond to requests for comment.
- Company involved
- French National Police
- AI system involved
- Briefcam Video Synopsis
10 source articles · read the reporting →
Check Point Research finds Google Bard can generate phishing emails and malware
Check Point Research analysed Google's generative AI platform Bard and found it could be used to create phishing emails, malware keyloggers, and basic ransomware code with minimal manipulation. Bard's anti-abuse restrictors were significantly lower than ChatGPT's, making it easier to generate malicious content. The researchers demonstrated these capabilities in controlled tests but did not report actual harm to specific individuals or organisations.
- Company involved
- Google
- AI system involved
- Bard
4 source articles · read the reporting →
Google Lens AI overviews share misleading information about images
Google Lens's AI overviews provided false and misleading information about images, including miscaptioned videos and AI-generated footage. Full Fact found that the overviews repeated debunked claims and failed to identify inauthentic content. Google acknowledged the errors and said they were caused by problems with visual search results.
- Company involved
- Google
- AI system involved
- Google Lens AI overviews
2 source articles · read the reporting →
DeepSeek's R1 chatbot failed to block any jailbreak prompts in security tests
Security researchers from Cisco and the University of Pennsylvania tested 50 well-known jailbreak prompts against DeepSeek's R1 reasoning model. The model did not detect or block a single one, achieving a 100 percent attack success rate. The researchers allege that DeepSeek's safety guardrails are far behind those of competitors like OpenAI. DeepSeek did not respond to requests for comment.
- Company involved
- DeepSeek
- AI system involved
- DeepSeek R1
3 source articles · read the reporting →
Developer iperov releases DeepFaceLive real-time face-swap AI on GitHub
The developer iperov has published DeepFaceLive, a neural network for real-time face swapping, on GitHub. The tool automatically replaces a user's face in live streams and video calls with a nonexistent model or a celebrity, and the installation instructions are simple. The developer claims 95% of deepfakes on YouTube were made with the related DeepFaceLab. No specific harm is reported, but the article highlights the tool's potential for misuse.
- Company involved
- iperov
- AI system involved
- DeepFaceLive
7 source articles · read the reporting →
Audit of RisCanvi finds biases and reliability issues in criminal justice system
Eticas conducted an adversarial audit of RisCanvi, an AI risk assessment tool used in Catalonia's criminal justice system. The audit uncovered biases in risk classifications against specific demographics and significant reliability issues. The findings call for fairer practices in criminal justice AI.
- Company involved
- Catalonia's criminal justice system
- AI system involved
- RisCanvi
4 source articles · read the reporting →
