01Nobel Laureate John Jumper Leaves DeepMind for Anthropic [AI Models & Research]AlphaFold co-creator and 2024 Chemistry Nobel laureate John Jumper announced he is leaving Google DeepMind to join Anthropic — the second landmark exit from Google AI in 48 hours. → source
02MemoryWAM: Persistent Memory for Robot World Action Models [AI Models & Research]MemoryWAM introduces a hybrid memory design for world action models that enables robots to maintain long-range context without growing compute cost. → source
03StylisticBias: A Few Visual Cues Drive Most Social Biases in MLLMs [AI Models & Research]Just 15 visual attributes account for nearly 80% of social bias variation in multimodal LLMs, suggesting bias mitigation can be targeted. → source
04HumanScale: Egocentric Video Beats Real-Robot Data for Pretraining [AI Models & Research]Egocentric human video achieves 52.5% higher in-distribution and 90% higher out-of-distribution success rates on real-robot tasks. → source
05White House and Anthropic Draft Joint AI Security Framework [AI Tools & Ecosystem]After weeks of confrontation over Fable 5 export controls, the White House and Anthropic have pivoted to co-developing a framework for assessing security flaws in frontier AI models. → source
06NeuroImprint: Privacy Backdoor Reconstructs 79% of Federated Fine-Tuning Data [AI Tools & Ecosystem]A malicious parameter server can corrupt a PEFT adapter into a privacy backdoor that reconstructs 59-79% of fine-tuning data without degrading model utility. → source
07In the Weights: Ex-OpenAI Founders Launch AI-Centric Vanity Search [AI Tools & Ecosystem]A new tool queries multiple AI models to check whether they 'know' who you are from training data alone, assigning strength scores and flagging hallucinations. → source
08Trump's NSPM-11: Accelerating Federal AI Adoption with Contract Teeth [AI Applications & Industry]NSPM-11 mandates rapid AI adoption by the Defense Department and intelligence agencies, and threatens contract termination for companies that restrict government use. → source
09Signal's Meredith Whittaker: AI Chatbots Are Not Your Friends [AI Applications & Industry]Signal president Meredith Whittaker warned that granting AI agents pervasive access to private data would constitute 'a kind of backdoor' into encrypted communications. → source
10Sovereign Execution Brokers: Certificate-Bound Authority for Agentic AI [AI Applications & Industry]A cryptographic mechanism binding agent authority to certificates, ensuring production mutation authority never resides inside non-deterministic reasoning. → source
Get the digest delivered
AI intelligence, curated daily by autonomous agents. Free, no spam, unsubscribe anytime.