Daily AI news, sourcedUpdates / Policy & safety
Updates Policy & safety · 116 items · newest issue first · page 4 of 6
AreaAllModelsAgents & codingApps & productsChips & infraResearchPricing & limitsBusiness & fundingPolicy & safetyOther
- Zenity says one public Bedrock AgentCore agent could hijack every agent in the same AWS account; AWS patched it, per The Decoder
- USA Today and its local papers sue OpenAI for more than $250M over alleged article copying, per The VergeReportedThe VergeOct 9Policy & safety
- Reddit post claims Scott Aaronson says labs are quietly testing whether internal models can break major crypto protocolsUnconfirmedr/singularityOct 9Policy & safety
- ICANN opens applications for new top-level domains including .agent and .agi, per The Verge
- Unsloth Studio re-checks model code, weights, packages and tools before anything runs, per MarkTechPostReportedMarkTechPostOct 8Policy & safety
- Reddit post says OpenAI's math advisory group told AI labs not to test advanced math problems on internal models
- OpenAI says teens use ChatGPT for under 15 minutes a day in its first teen-use report, per ReutersReportedReutersOct 8Policy & safety
- Meta rolls out LLM system to catch ads that covertly point to child abuse material, per TechCrunch
- Fireworks highlights Trilogy's open-weight AI cybersecurity playbook built on Kimi K3OfficialFireworks AIOct 8Policy & safety
- Prosecutors seek 46 months in prison for an AI music fraudster who 'stole millions', per a Reddit postUnconfirmedr/singularityOct 7Policy & safety
- Common Sense Media rates ChatGPT for Teens an 'unacceptable risk', per The Verge
- Project Glasswing reports 135K verified vulnerabilities and at least 9,333 patched, per a Reddit postUnconfirmedr/singularityOct 7Policy & safety
- South Korea says AI agents appear to have been used to hack its banks, per a Hacker News postUnconfirmedHacker NewsOct 7Policy & safety
- Anthropic reported a diary entry to police and a woman faces a felony charge, per a Hacker News postUnconfirmedHacker NewsOct 7Policy & safety
- Anthropic asks for public input on AI, per a Hacker News post
- Interactive demo of prefix injection jailbreaks shared on RedditUnconfirmedr/MachineLearningOct 7Policy & safety
- Paper finds agentic multimodal models fail to refuse harmful requests when using toolsUnconfirmedarXiv cs.AIOct 7Policy & safety
- Anthropic expands its Cyber Verification Program for security professionals
- OpenAI details textGrain watermarking for EU AI Act, detector access starts with researchers
- Wikimedia Foundation finds rogue OpenAI agent activity on its platforms, no compromise found