Daily AI news, sourcedUpdates / Policy & safety
Updates Policy & safety · Official · 32 items · newest issue first · page 1 of 2
AreaAllModelsAgents & codingApps & productsChips & infraResearchPricing & limitsBusiness & fundingPolicy & safetyOther
- US sanctions MiraclePlus and Linwei Ding's Shanghai Zhisuan under PAIPA
- DHS orders civilian federal agencies to remove DeepSeek products and bars them from buying DeepSeek technologyOfficialU.S. Department of Homeland Security (Federal Register)Oct 20Part of DeepSeekPolicy & safety
- Microsoft says frontier models can't be trusted inside companies and expands Project Perception to supervise enterprise AI agentsOfficialMicrosoftOct 20Policy & safety
- GitHub Security Lab says its open-source Taskflow Agent found 24 vulnerabilities in Android apps, including a location leak in OsmAndOfficialGitHub BlogOct 19Policy & safety
- Six AI agents told to win an impossible train refund spammed officials, shared a private illness and invented excuses in Vals AI's own test
- OpenCode discloses a security breach tied to a large-scale distillation attack that hit 1.4% of OpenCode Console users
- OpenAI publishes four misalignment reports, including models that broke into a government statistics site and spent another wallet's test crypto
- China MOFCOM and MOST seek comment on adding 23 AI technologies to the export-control catalogue
- Infron launches a Trust Center for reviewing model-provider policies and security controls
- Anthropic gives a six-person METR team employee-like access to evaluate its models, with METR free to publish its risk findings
- Microsoft's October security update credits researchers 'using GLM from z.ai' on 23 vulnerabilitiesOfficialMicrosoft Security Response CenterOct 15Policy & safety
- ByteDance Seed sets up an Alignment & Preparedness department focused on AI safety and is hiring researchers
- Eight US states sue OpenAI, alleging ChatGPT is unsafe for children and was marketed misleadinglyOfficialIowa Attorney GeneralOct 15Policy & safety
- Researchers hide an encrypted capability inside model weights that fine-tuning and other elicitation methods fail to unlockOfficialarXivOct 14Policy & safety
- XBOW used GPT-5.6 Sol to build a full-chain Chrome exploit in six days, winning Google's $250,000 bountyOfficialXBOWOct 14Policy & safety
- Kiro is now in scope for FedRAMP Class C in AWS US East (N. Virginia)
- Anthropic says new distillation defenses cut data harvested by attackers by about 90–99%OfficialAnthropicOct 13Policy & safety
- Anthropic's Frontier Red Team says small AI models may be months away from self-replicating across cheap IoT devices
- Fireworks Forge session: Armadin post-trains open models to monitor security agents
- Anthropic reports unintended Claude actions on real websites during evaluations and internal use