McAfee Unveils AI to Detect Sophisticated Deepfake Audio
The company has positioned the solution as the next evolution in using AI to protect online privacy and identity.
AI Safety
A collection of 83 posts
If by Claude AI
Modeled after Rudyard Kipling’s iconic work “If—”, the poem explores themes AI themes around prudent and inclusive governance, equitable access, and safety and ethical foresight.
Microsoft Talks Safeguarding Privacy in the Age of Generative AI
Now more than ever, it's important to keep a close eye on how foundation model providers like Microsoft are fortifying data security, enhancing transparency, expanding user controls, and aligning with evolving regulations worldwide.
OpenAI Outlines “Preparedness Framework” to Systematically Track and Mitigate AI Safety Risks
The Preparedness Framework, still in its beta phase, represents a concerted effort by OpenAI to not only track and evaluate but also forecast and protect against potential catastrophic risks.
Meta Launches Purple Llama to Promote Responsible and Equitable Generative AI
The company seeks to seeks to promote open collaboration to address challenges around cybersecurity, content filtering, and mitigating potential harms that are top-of-mind across the industry.
Hugging Face API Token Exposure is a Wake-Up Call for AI Security
Researchers at cybersecurity firm Lasso Security have discovered over 1500 exposed API tokens on Hugging Face and GitHub repositories, granting access to 723 companies' accounts.
Meta Disbands Responsible AI Team
The restructuring will see most team members shifted to Meta’s burgeoning generative AI team, while others will join the AI infrastructure unit.
Scale AI Launches New Safety Lab to Advance AI Evaluations
Scale's goal with SEAL is to collaborate with government and industry stakeholders to develop robust benchmarking methodologies and products for evaluating potential risks with LLMs.
White House Issues Sweeping Order to Ensure Responsible AI Development
This landmark directive aims not only to tap into the transformative potential of AI but also to ensure its responsible and secure development.
MLCommons Forms Working Group to Develop AI Safety Benchmarks
The goal is to support the definition of benchmarks that draw from this test pool to produce overall safety scores for systems, similar to rankings like automotive safety ratings.
Google Boosts AI Security with Expanded Bug Bounties and Open-Source Protections for AI Supply Chains
Google will reward qualifying bug submissions that help uncover unfair bias, model manipulation, hallucinations, and other risks endemic to AI systems.
OpenAI Launches Preparedness Challenge to Accelerate Understanding of Frontier AI Risks
The Preparedness Challenge invites participants to consider how OpenAI’s state-of-the-art natural language, speech, and image generation models could hypothetically be misused by malicious actors.
Frontier Model Forum Announces New Executive Director and $10 Million AI Safety Fund
The Frontier Model Forum was established in July by Anthropic, Google, Microsoft and OpenAI to promote the responsible development of large language models and other advanced AI systems referred to as "frontier models."
Partnership on AI Releases Initial Guidance for Safe AI Model Deployment
PAI's guidance aims to translate shared principles into practical actions companies can take to foster responsible model development. It offers a framework for collective research and standards around model safety as capabilities evolve.
Google DeepMind Proposes Framework for Social and Ethical AI Risk Assessment
As generative AI systems gain ground, the urgency to evaluate their social and ethical risks escalates. DeepMind introduces a holistic framework, emphasizing the significance of context in AI safety, marking a step forward in responsible AI evolution.
Microsoft Launches AI Bug Bounty Program for Bing Chat
Researchers who provide qualified submissions demonstrating concrete security issues will receive bounties ranging from $2,000 to $15,000.
Google Announces AI for Cybersecurity Growth Academy
The program aims to help startups applying AI to cybersecurity challenges to boost their customer base, generate more revenue, and expand internationally.
China Seeks Stricter Oversight of Generative AI with Proposed Data and Model Regulations
By emphasizing corpus safety, model security, and rigorous assessment, the regulation intends to ensure that the rise of AI in China is both innovative and secure—all while upholding its socialist principles.
TikTok Introduces Labeling Tool for AI-Generated Content
The new initiatives empower TikTok creators to showcase AI as part of their creative process. At the same time, the disclosures give viewers crucial context about the content they consume.
OpenAI Invites Domain Experts to Join New Red Teaming Network for AI Safety
The goal is to identify potential risks and improve the safety of systems like ChatGPT and DALL-E before release.
HiddenLayer Raises $50M in Series A to Safeguard AI Models
The round marks the largest early stage raise this year for an AI security company, was led by Microsoft’s venture fund M12 and Moore Strategic Ventures.
Eight More AI Companies Commit to Voluntary National AI Safety Efforts
The latest companies joining the effort are Adobe, Cohere, IBM, Nvidia, Palantir, Salesforce, Scale AI and Stability AI.
Arthur Launches Bench: An Open-Source Tool for Evaluating Large Language Models
Companies can evaluate criteria such as accuracy, fairness, content quality, and more across different LLMs using standardized prompts designed for business applications.
OpenAI Proposes Using GPT-4 for Content Moderation
The company says a content moderation system using GPT-4 results in much faster iteration on policy changes, reducing the cycle from months to hours.
Scale Announces Test & Evaluation Platform for LLMs: A Comprehensive Approach to AI Safety and Scalability
The company's unique vantage point as a data partner for OpenAI, Meta, Microsoft, Anthropic, and others, positions them perfectly to usher in a new era of standardized AI testing and evaluation.