CyberSecurity updates
Updated: 2024-11-10 13:30:23 Pacfic


Mistral
Mistral AI Launches Automatic Content Moderation Tool - 1d

French tech company Mistral AI has launched a new AI-powered content moderation tool that can detect and remove offensive or illegal posts automatically. The tool, based on the Ministral 8B AI model, aims to improve content moderation on social media platforms by automatically flagging and removing inappropriate content. However, there is still a risk of misjudgments, as studies have shown that posts about people with disabilities can be flagged as “negative” or “toxic” even though that’s not the case. The tool will initially support Arabic, English, French, Italian, Japanese, Chinese, Korean, Portuguese, Russian, Spanish and German, with more languages ​​on the way.

Devin Coldewey @ TechCrunch
ChatGPT and DALL-E AI Safety Measures During Election - 1d

During the recent US elections, OpenAI took measures to mitigate potential risks associated with their AI tools. ChatGPT was programmed to steer users away from seeking election news on November 5th and 6th, effectively limiting its role in disseminating election information. Additionally, DALL-E was configured to reject requests for images of Trump and other individuals, preventing the creation of potentially harmful deepfakes or misinformation. These actions demonstrate OpenAI’s commitment to responsible AI development and their efforts to prevent their tools from being used to spread disinformation or influence elections.

Benj Edwards @ Ars Technica
Mistral AI Launches Content Moderation Tool - 1d

French tech company Mistral AI has launched a content moderation tool powered by its large language model, Ministral 8B, capable of automatically detecting and removing offensive or illegal content from online platforms. This tool is designed to help businesses and organizations mitigate risks associated with harmful content by detecting and filtering out problematic posts. Mistral’s moderation tool is initially available for Arabic, English, French, Italian, Japanese, Chinese, Korean, Portuguese, Russian, Spanish, and German, with more languages planned for the future.

Devin Coldewey @ TechCrunch
ChatGPT Moderates Election News and Deepfakes - 15h

OpenAI’s ChatGPT chatbot actively moderated election news and deepfake requests during the recent US election. It provided responses asking users to seek election news elsewhere on November 5th and 6th, rejecting over 2 million requests. Additionally, ChatGPT refused to generate DALL-E images depicting individuals like Donald Trump, blocking over 250,000 requests. This proactive measure aimed to curtail misinformation and the potential spread of deepfakes, reflecting a conscious effort to combat the influence of AI-generated content in elections.

cnbc.com
Tesla Warned for Misleading Claims About Autonomous Vehicles - 1d

The National Highway Traffic Safety Administration (NHTSA) warned Tesla in May regarding misleading social media posts that falsely suggested their vehicles are autonomous. These claims could potentially endanger drivers by giving them a false sense of security and leading to misinterpretations of the car’s capabilities. The NHTSA’s warning highlights the importance of clear communication and avoiding misleading claims regarding autonomous vehicles. This incident underscores the need for responsible development and deployment of autonomous vehicle technology, focusing on safety and avoiding potential misinterpretations that could endanger lives.

Flag this
SPAR Program for AI Safety and Governance Research - 4d

The SPAR program is a remote-first part-time program connecting mentors and mentees for three-month AI safety and governance research projects. This initiative aims to promote research in areas such as technical AI safety, AI policy and governance, AI strategy, and AI security. The program offers funding for compute costs and provides valuable experience for both mentors and mentees. The focus on AI safety and governance is increasingly relevant in the rapidly evolving landscape of artificial intelligence. The SPAR program plays a crucial role in fostering research and development within the AI community.

Bruce Schneier @ Security Boulevard
AI Discovering Vulnerabilities: A New Era of Security - 4d

The use of Artificial Intelligence (AI) to automatically discover vulnerabilities in code is becoming increasingly prevalent, with researchers developing new methods to effectively scan source code and find zero-days in the wild. Companies like ZeroPath are combining deep program analysis with adversarial AI agents to uncover critical vulnerabilities, often in production systems, that traditional security tools struggle to detect. While AI-based vulnerability discovery is still in its early stages, its potential to enhance security measures is undeniable. This development could significantly improve the effectiveness of security testing and lead to the identification of vulnerabilities earlier in the development cycle, reducing the risk of exploitation.

reuters.com
Meta Opens Llama AI for US National Security - 5d

Meta has opened up its open-source Llama AI models to US government agencies and contractors for use in national security applications. This move aims to enhance the US’s capabilities in areas such as logistics, cyber defense, and counterterrorism efforts. The decision comes amidst concerns about China’s rapid advancements in AI and the potential threat posed by its military AI development. Meta is collaborating with companies like Amazon, Microsoft, and Lockheed Martin to make Llama accessible to the government, emphasizing the importance of American AI dominance in the global AI race.

Igor Bonifacic @ Engadget is a web magazine with obsessive daily coverage of everything new in gadgets and consumer e
Zero-Click Vulnerability Found in Synology Photos App - 8d

Security researchers discovered a critical zero-click vulnerability within Synology’s Photos app, impacting millions of users. This vulnerability allows attackers to compromise the system without any user interaction. The flaw resides in a part of the app that doesn’t require authentication, enabling direct exploitation over the internet. Attackers can gain root access and install malicious code, potentially turning the infected device into a botnet for further nefarious activities. Synology has addressed the bug, but users need to manually update their devices. This incident highlights the importance of regularly updating software to mitigate security risks and the growing threat of zero-click vulnerabilities.

Tom Warren @ The Verge
Windows Recall Feature Delayed Again - 9d

Microsoft’s ambitious AI-powered Windows Recall feature, which was originally slated for release in October, has been delayed again until December. The feature, designed to create searchable snapshots of user activity, has faced criticism over its potential for privacy violations and security concerns. Despite these concerns, Microsoft is committed to delivering Recall, but is taking additional time to ensure a “secure and trusted experience.” The company has made efforts to address privacy worries, including making Recall an opt-in feature and emphasizing enhanced security measures. However, the repeated delays suggest that the development process is proving more challenging than anticipated.

opensource.org
Open Source AI Definition Sparks Debate - 12d

The Open Source Initiative (OSI) has released a new definition of open-source artificial intelligence, which includes a requirement for AI systems to disclose their training data. This definition directly challenges Meta’s Llama, a popular open-source AI model that does not provide access to its training data. Meta has argued that there is no single open-source AI definition and that providing access to training data could pose safety concerns and hinder its competitive advantage. However, critics argue that Meta is using these justifications to minimize its legal liability and protect its intellectual property. OSI’s definition has been met with support from other organizations like the Linux Foundation, which are also working to define open-source AI. The debate highlights the evolving landscape of open-source AI and the potential conflicts between transparency, safety, and commercial interests.

universalmusic.com
Universal Music Partners with Klay AI to Create Ethical Music Generator - 12d

Universal Music Group (UMG) has entered into a partnership with Klay Vision, an AI company focused on ethical music generation, to develop a “Large Music Model” named KLayMM. The model aims to leverage AI for music creation while respecting copyright and artist rights, promoting collaboration between human artists and AI technology. The partnership emphasizes ethical AI development, ensuring copyright and artist rights are protected, and creating opportunities for new creative avenues. The collaboration seeks to address concerns about copyright infringement and artist exploitation often associated with AI-generated music. It highlights a growing trend of music companies embracing AI technology, with UMG also involved in various lawsuits and licensing arrangements related to AI music.

learn.microsoft.com
AI-Powered Medical Transcription Tool 'Nabla' Faces Concerns Over Hallucinations - 12d

Concerns have arisen regarding the accuracy of an AI-powered medical transcription tool called ‘Nabla’, which is being used by medical professionals. Researchers have discovered that the tool, powered by OpenAI’s Whisper AI, exhibits a significant hallucination rate, fabricating information that was not spoken in the original recordings. This issue is particularly concerning in healthcare, where the accuracy of medical records is paramount. The hallucination rate, estimated to be around 1-2%, means that a considerable number of transcribed sentences contain fabricated information. Although individual users may not notice these errors, they could have catastrophic consequences at scale, potentially leading to incorrect diagnoses, inappropriate treatments, and even harm to patients. This highlights the critical need for robust safeguards and verification measures to ensure the reliability of AI-powered medical transcription tools.

Fi, infosec
AI Chatbots and Data Summarization: A Potential Threat to Employment - 11d

The use of AI chatbots for summarizing emails and other messages raises concerns about its potential impact on employment. If AI can efficiently process and summarize business communications, it might render human employees in these roles redundant. The argument suggests that AI can handle these tasks with greater accuracy and efficiency, potentially reducing the need for human intervention. This raises questions about the future of human roles in data analysis and information processing, particularly in fields like business communication and customer service. The increased adoption of AI for these tasks necessitates a reevaluation of the role of humans in these areas and the need for adaptation and reskilling in order to remain relevant in the evolving job market.

Jeremy Gray @ PetaPixel
Apple Intelligence Launches, Integrates AI Across iPhone, iPad, and Mac - 12d

Apple has launched its new AI-powered system, Apple Intelligence, integrating it across iPhone, iPad, and Mac devices. This update brings enhanced features including an improved Siri, writing tools, and a Clean Up function for photos. The Clean Up tool leverages AI to remove distractions from images while maintaining realism. This move demonstrates Apple’s commitment to incorporating AI into its product ecosystem, offering users a seamless and integrated experience across its devices. While initially available only in English, Apple has confirmed that support for additional languages will be added in the future, further expanding the reach of this new AI platform.


This site is an experimental news aggregator using feeds I personally follow. You can reach me at Bluesky if you have feedback or comments.