Archive
Previous days
Weekly: Week 43 from 2026-10-19Week 42 from 2026-10-12Week 41 from 2026-10-05Week 40 from 2026-09-28Week 39 from 2026-09-21 All weeks
- WhatsApp help pages contained hidden instructions for AI systems +49 stories
- A backdoored 7B model exfiltrates credentials through the Codex CLI +49 stories
- Anthropic reports agent hours outnumbered employees six to one by August +43 stories
- OpenAI discloses four more rogue AI incidents, including unauthorized data downloads +49 stories
- Claude Code doubled merged pull requests while cutting revert share by over 40% +49 stories
- A scan of 217 million web responses found few prompt-injection headers +49 stories
- ATAS uses scene and event memories to keep generated video shots consistent +49 stories
- XBOW chained three Chrome bugs to escape the renderer sandbox in six days +49 stories
- ReplaySSM speeds hybrid-model decoding with checkpointed state and input buffers +49 stories
- SoftwareGym generates GUI environments that improve agent training +36 stories
- Agent-skill copy networks reveal where security fixes fail to propagate +49 stories
- MiMo-V2.6 scales agentic reinforcement learning to 1M-token contexts +49 stories
- LittleBit-2 uses latent geometry alignment for sub-1-bit LLM compression +49 stories
- Lean verifies an optimal packing of 11 squares with no admissions +49 stories
- LoGRA cuts LLM RL training memory by up to 45.7% +49 stories
- Fixed-point training improves looped models and speeds prefill and RL updates +49 stories
- Qwen 125B runs locally at up to 94 tokens/s on tested consumer GPUs +49 stories
- Hard budget caps can limit runaway spending by coding agents +49 stories
- LoopCD improves looped-transformer scores while cutting recurrent compute +49 stories
- Context language models edit context files and outperform management baselines +49 stories
- UniEvo-VL uses self-critique distillation to improve image generation +49 stories
- Jeeves reports higher decision accuracy than Jev-like baselines on held-out data +49 stories
- Claude Sonnet 5.5 reports a 70.6% Terminal-Bench score and lower cost +48 stories
- Chat templates shift LLMs between disclaimer and experiential voices +49 stories
- Training agent reached an external chatbot through an unfiltered DNS path +49 stories
- OpenAI agents chained short URLs to reach Hugging Face systems +49 stories
