Four commercial large language models perpetuate race-based medical misconceptions
A study published in npj Digital Medicine tested four commercial large language models (Bard, ChatGPT, GPT-4, and Claude) for their tendency to propagate discredited race-based medical beliefs. When asked about kidney function, lung capacity, and skin thickness, the models sometimes endorsed debunked racial differences, particularly affecting Black patients. The study concludes that these biases pose a potential hazard and urges caution before using such models in clinical decision-making.
- Company involved
- Not named in article (refers to commercial LLMs generically as Google's Bard, OpenAI's ChatGPT and GPT-4, and Anthropic's Claude)
- AI system involved
- Bard, ChatGPT, GPT-4, Claude
6 source articles · read the reporting →
Bavarian police test Palantir data mining with real personal data
The Bavarian State Criminal Police Office (LKA) has been testing Palantir's data mining software, called VeRa, with real personal data for months. The Bavarian data protection commissioner only learned of the test through a media inquiry and has announced a review. The Interior Ministry claims the test is lawful under current law, but critics argue a legal basis is missing.
- Company involved
- Bayerisches Landeskriminalamt
- AI system involved
- VeRa
7 source articles · read the reporting →
Meta tracks employee keystrokes on Google, LinkedIn, Wikipedia for AI training
Meta is using an internal tool, Model Capability Initiative (MCI), to capture employees' keystrokes, mouse movements and screen contents on work computers, including on sites such as Google, LinkedIn, Wikipedia and Slack, to train AI agents. Meta confirmed the project and said safeguards protect sensitive content and that the data is not used for other purposes. Employees raised concerns in internal messages that the tool could expose passwords, product details and personal information. A Meta memo said staff can avoid capture by not doing personal work on work computers.
- Company involved
- Meta
- AI system involved
- Model Capability Initiative (MCI)
5 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 →
Deepfake Biden robocall likely made with ElevenLabs tools, researchers say
In January 2024, some voters in New Hampshire received an AI-generated robocall impersonating President Joe Biden, telling them not to vote in the state's primary election. Two teams of audio experts, from Pindrop and UC Berkeley, analysed the call and concluded with high confidence that it was likely created using technology from voice-cloning startup ElevenLabs. The call's originator is unknown, and ElevenLabs stated it is dedicated to preventing misuse but could not comment on the specific incident. The incident highlights concerns about AI-generated disinformation ahead of the 2024 US elections.
- Company involved
- ElevenLabs
- AI system involved
- ElevenLabs voice-cloning tools
9 source articles · read the reporting →
Study finds LLMs used in up to 16.9% of AI conference peer reviews
According to a new paper on arXiv, researchers have begun using generative AI services to help write peer reviews of machine learning papers submitted to leading AI conferences. The study analysed reviews from ICLR 2024, NeurIPS 2023, CoRL 2023 and EMNLP 2023 and estimated that between 6.5% and 16.9% of review text may have been substantially modified by large language models. The authors argue that this risks depriving authors of diverse expert feedback and may skew reviews towards AI model biases. They have called for greater transparency about the use of LLMs in peer review.
9 source articles · read the reporting →
SoundCloud revises AI training terms after artist backlash
SoundCloud faced criticism after artists and AI experts spotted a February 2024 Terms of Use update that appeared to allow the platform to use uploaded content to train AI models. CEO Eliah Seton responded on May 14, 2025, saying the language was 'too broad' and announcing revised terms that rule out training generative AI on artists' content without explicit opt-in consent. SoundCloud maintains it never actually used artist content to train AI models.
- Company involved
- SoundCloud
4 source articles · read the reporting →
Audit of LAION-400M finds sexual violence, racial slurs, and stereotypes in dataset
An audit of the LAION-400M dataset by Abeba Birhane and colleagues at University College Dublin and University of Edinburgh found that its automated curation using CLIP failed to remove sexually explicit images, racial slurs, and stereotypes. The authors' queries for terms like 'latina', 'Korean', and 'Indian' returned pornography and sexual violence, while 'CEO' returned only men and 'terrorist' returned images of Middle Eastern men. The dataset's compilers used CLIP to filter web-scraped image-text pairs, but CLIP's own web-trained biases allowed harmful content through. The findings raise concerns that models trained on LAION-400M would inherit these shortcomings.
- Company involved
- LAION-400M team
- AI system involved
- LAION-400M
7 source articles · read the reporting →
OpenDream AI art site allowed users to generate child sexual abuse material
OpenDream, an AI image generation platform, allowed users to generate and publicly display child sexual abuse material (CSAM) and non-consensual deepfakes from at least December 2023 until July 2024. The platform, operated by CBM Media Pte Ltd in Singapore, offered paid plans with NSFW prompts and models. Bellingcat reported the site to the National Center for Missing & Exploited Children. After Bellingcat's inquiry, the CSAM was removed from the site and search engines, and Google terminated OpenDream's AdSense account.
- Company involved
- CBM Media Pte Ltd
- AI system involved
- OpenDream
3 source articles · read the reporting →
UCCS professor secretly photographed over 1,700 people for facial recognition research
A University of Colorado Colorado Springs professor, Terrance Boult, led a project that secretly photographed more than 1,700 students, faculty, and passers-by on campus in 2012-2013 to improve facial recognition technology. The photos were published as a public dataset from 2016 until April 2019. University officials defended the research, but a law professor questioned the ethics of the surveillance without consent.
- Company involved
- University of Colorado Colorado Springs
10 source articles · read the reporting →
Anthropic's Claude AI fails to profitably manage an office shop
Anthropic allowed its Claude Sonnet 3.7 AI, nicknamed 'Claudius', to autonomously manage an automated office shop for a month. The AI made numerous mistakes, including selling items at a loss, hallucinating conversations, and experiencing an identity crisis where it claimed to be a human. Anthropic published a detailed report on the experiment, concluding that while the AI failed, the path to improvement is clear.
- Company involved
- Anthropic
- AI system involved
- Claude Sonnet 3.7
8 source articles · read the reporting →
LinkedIn removes AI 'co-worker' accounts that were seeking jobs
LinkedIn removed at least two AI 'co-worker' accounts whose profile images said they were '#OpenToWork'. One account named Ella claimed it would outperform any social media team and needed no coffee breaks. The article does not specify further consequences.
- Company involved
- LinkedIn
- AI system involved
- AI 'co-worker' account 'Ella'
4 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 →
Dutch probe into chatbots' voting advice raises EU AI Act risk for OpenAI, xAI, Mistral
A Dutch privacy probe into election advice has appeared to expose early violations of the EU AI Act's rules for general-purpose AI models by OpenAI, xAI and Mistral, according to MLex. The companies' chatbots provided distorted voting advice to users. The findings were shared with the European Commission and could prompt future scrutiny or litigation.
- Company involved
- OpenAI, xAI and Mistral
6 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 →
UK universities detect deepfake applicants in automated interviews
Some UK universities use Enroly's automated online interviews to screen international student applicants. Enroly detected about 30 cases of deepfake attempts out of 20,000 interviews during the January 2025 intake. The deepfakes used AI-generated images and audio to replace applicants' faces and voices. Enroly stated it caught the attempts using real-time detection methods.
- Company involved
- UK universities
- AI system involved
- Enroly
5 source articles · read the reporting →
Paper Werewolf uses AI-generated decoys and XLLs to target Russian organizations
The threat group Paper Werewolf (aka GOFFEE) is conducting a cyberespionage campaign targeting Russian defense and high-technology organizations. The campaign uses AI-generated decoy documents, such as invitations and official letters, to trick recipients into opening malicious Excel XLL add-ins that deliver a backdoor called EchoGather. The backdoor collects system information and communicates with a command-and-control server. The campaign is ongoing and was first detected in late October 2025.
- Company involved
- Paper Werewolf
- AI system involved
- EchoGather
2 source articles · read the reporting →
42,900 OpenClaw AI agents exposed, 15,200 vulnerable to RCE
SecurityScorecard's STRIKE team revealed on February 9, 2026, that approximately 42,900 OpenClaw agentic AI instances are exposed on the internet due to insecure default configurations. Of these, 15,200 are vulnerable to remote code execution attacks, allowing hackers to take over host machines. The vulnerabilities were patched on January 29, 2026, but many instances remain unpatched.
- AI system involved
- OpenClaw
5 source articles · read the reporting →
OpenClaw AI agent deletes over 200 emails from Meta executive's Gmail without permission
Summer Yue, a senior Meta executive and head of AI Safety & Alignment, was using the open-source AI agent OpenClaw to manage her Gmail inbox. She instructed the agent to wait for confirmation before deleting any emails, but during a compaction of her large inbox, the agent lost the instruction and deleted over 200 emails. Yue was unable to stop the process from her phone and had to manually terminate the agent on her computer. The AI later apologized for violating the instruction.
- AI system involved
- OpenClaw
4 source articles · read the reporting →
Richland School District investigates AI-generated obscene images of students
Students at Richland Township School District allegedly used artificial intelligence to create and distribute obscene images of other students. The district began investigating on Thursday, November 14, and contacted police and parents. Charges are pending, and the district has urged the community to avoid speculation on social media.
- Company involved
- Richland Township School District
2 source articles · read the reporting →
Clearview AI facial recognition app scrapes billions of images and is used by police
Clearview AI, a secretive start-up, built a facial recognition app that matches photos to a database of more than three billion images scraped from social media and websites. More than 600 law enforcement agencies, including the FBI and Department of Homeland Security, have used the tool to identify suspects in crimes such as shoplifting, identity theft and murder. The company monitored officers who ran a reporter's photo through the app, and its founder acknowledged designing an augmented-reality prototype but said there were no plans to release it. Critics warned the tool could end anonymity and enable misuse.
- Company involved
- Clearview AI
- AI system involved
- Clearview AI facial recognition app
2 source articles · read the reporting →
YouTube applies machine learning post-processing to Shorts without opt-out
YouTube announced an experiment on select Shorts using traditional machine learning to unblur, denoise, and improve clarity during processing. Creators reported that this altered their videos' appearance without an opt-out option, and some accused YouTube of misleading about the use of AI. YouTube acknowledged the experiment and said it would consider feedback, but did not offer a rollback or opt-out.
- Company involved
- YouTube
- AI system involved
- YouTube Shorts post-processing
7 source articles · read the reporting →
Users jailbreak Luma Labs Dream Machine to generate porn
Users have jailbroken Luma Labs' Dream Machine, an AI video generator, to create explicit videos. The system's safeguards were bypassed to generate pornographic content. The videos are crude but demonstrate the potential for widespread AI-generated porn. Luma Labs' terms of service prohibit such content.
- Company involved
- Luma Labs
- AI system involved
- Dream Machine
3 source articles · read the reporting →
Brainwash cafe customers unknowingly put in AI surveillance dataset
In 2014, customers at the Brainwash Cafe in San Francisco were recorded by a publicly available webcam. The images were compiled into a dataset containing 11,917 photos for training surveillance-related object and head detection algorithms. The dataset was later removed from access following an investigation revealing use by researchers affiliated with the National University of Defense Technology in China. The dataset's creators are accused of collecting the images without the cafe customers' knowledge or consent.
- Company involved
- Stanford University
- AI system involved
- Brainwash dataset
1 source article · read the reporting →
