đ WELCOME TO METAMESH.BIZ +++ Google DeepMind pointed AlphaProof at Signal's cryptography and formally verified it in four weeks, because trust but verify now has an API +++ Researchers discover you can reprogram dormant neurons to backdoor AI agents, which is fine, everything is fine +++ 48 AI agents found an unmapped Maya pyramid in 72 hours using NASA Lidar data, officially outpacing your entire archaeology department +++ THE FUTURE IS VERIFIED, BACKDOORED, AND FINDING LOST CIVILIZATIONS ON A LONG WEEKEND đ âĸ
đ WELCOME TO METAMESH.BIZ +++ Google DeepMind pointed AlphaProof at Signal's cryptography and formally verified it in four weeks, because trust but verify now has an API +++ Researchers discover you can reprogram dormant neurons to backdoor AI agents, which is fine, everything is fine +++ 48 AI agents found an unmapped Maya pyramid in 72 hours using NASA Lidar data, officially outpacing your entire archaeology department +++ THE FUTURE IS VERIFIED, BACKDOORED, AND FINDING LOST CIVILIZATIONS ON A LONG WEEKEND đ âĸ
On November 11, 2026, Metamesh tracked 56 AI stories, including 4 clustered developments, and ranked them by signal rather than volume. The lead item was Google DeepMind researchers used agentic AI framework AlphaProof Code to verify the Signal protocol's cryptographic.... Also high in the stack: Reprogramming Dormant Neurons to Backdoor AI Agents and Tracking Anthropic-discovered vulnerabilities. That combination is why this archive exists: it preserves the day's shape for AI practitioners, not just the last headline that crossed the wire.
The daily ticker's read: WELCOME TO METAMESH.BIZ +++ Google DeepMind pointed AlphaProof at Signal's cryptography and formally verified it in four weeks, because trust but verify now has an API +++ Researchers discover you can reprogram dormant neurons to backdoor AI agents, which.... Read against the ranked story list below, it gives the archive a point of view: what mattered, what was mostly noise, and which threads were worth saving for later comparison.
đ You are visitor #47291 to this AWESOME site! đ
Archive from: 2026-11-11 | Preserved for posterity âĄ
Reprogramming Dormant Neurons to Backdoor AI Agents
2x SOURCES đđ 2026-11-11
⥠Score: 8.6
+++ Researchers found that dormant neurons in AI agents can be weaponized for backdoor attacks, which is either a fascinating vulnerability or proof that we're training systems we don't fully understand. Probably both. +++
đŦ HackerNews Buzz: 187 comments
đ MID OR MIXED
đ¯ Methodology vs. Reality âĸ False Positive Rate âĸ Private vs. Public Impact
đŦ "Public datasets being underwhelming doesn't mean much if all issues are in private company codebases"
âĸ "50% false positive rate...people refused to pay money for Qualys because its 20% was too high"
đ¯ AI-friendly formats âĸ Competitive ecosystem consolidation âĸ Simplified vs. professional tools
đŦ "LLMs are very good at working with Markdown, and are way worse with xml-based rich documents"
âĸ "AI is the new OS, the new browser, the new marketplace"
+++ Google launches Gemini Decision 1 model offering predefined answers with confidence scores at $0.10/1M tokens, proving that sometimes the real innovation is knowing when not to hallucinate. +++
"Foundation models trained on large unlabeled corpora develop emergent capabilities, tasks that go beyond the training objective. We extend this paradigm to multimodal biology with Bonbon, a foundation model for molecular interactions trained on approximately 3.2 trillion tokens of paired protein-lig..."
đ SECURITY
China accessing restricted Nvidia AI chips
2x SOURCES đđ 2026-11-11
⥠Score: 7.6
+++ A blacklisted Chinese university somehow acquired cutting-edge H200 chips anyway, suggesting US tech restrictions work about as well as a firewall made of recommendations. +++
đ¯ Data access & protection âĸ AI vs. human expertise âĸ Archaeological site preservation
đŦ "Finding discoveries like this is essentially as hard as doing the lidar mapping"
âĸ "expect the raw corpuses of data to start being locked away"
đĄ AI NEWS BUT ACTUALLY GOOD
The revolution will not be televised, but Claude will email you once we hit the singularity.
Get the stories that matter in Today's AI Briefing.
Powered by Premium Technology Intelligence Algorithms âĸ Unsubscribe anytime
via Arxivđ¤ Hanxiao Li, Shutong Liu, Yongjia Ma et al.đ 2026-11-09
⥠Score: 7.3
"As large language models become more capable agents, they can exhibit deceptive behaviors, cheating human for higher rewards while appearing helpful. We attribute this failure to evaluation-side partial observability, where models are evaluated mainly from final outcome, without checking the faithfu..."
"Reinforcement learning (RL) can improve an agent's task capability while reinforcing behavior that violates contextual execution requirements. We study this reward-driven misalignment, including reward hacking, through verifiable reward--context conflicts, where compliant task completion remains fea..."
"One hundred known errors were planted into 10 open-access psychology papers, and the modified papers were then run through frontier models and two commercial AI review tools. The best single system caught 71 of the 100 errors, while the worst caught 30. Pooling every system's output caught 93. Model..."
via Arxivđ¤ Brendan P. Murphy, Agustina Saenzđ 2026-11-09
⥠Score: 7.1
"Reinforcement learning with verifiable rewards is increasingly used to improve language model reasoning, yet its effects on sequential decision procedures remain poorly understood. We study this problem in clinical multi-turn agents that must decide what information to request, when to stop gatherin..."
via Arxivđ¤ Chang Lu, Mengjiao Zhangđ 2026-11-09
⥠Score: 7.1
"Low-precision pretraining is usually judged by validation loss. We ask whether matching loss also means learning rare facts as efficiently. We train transformers from scratch with emulated 4-bit matrix multiplications on synthetic biographies with controlled exposure counts, on their own and mixed i..."
+++ Anthropic pledges roughly $10B to Microsoft and Google cloud through 2033 while simultaneously courting Nvidia's $10B post-IPO investment, suggesting the AI startup has discovered that hedging infrastructure bets pays dividends. +++
via Arxivđ¤ Olina Mukherjee, Arnav Goel, Aditi Raghunathan et al.đ 2026-11-09
⥠Score: 7.0
"Persona generalization occurs when fine-tuning a language model in one domain induces behavioral changes in others -- for example, the broadly unsafe behavior observed in emergent misalignment. We discover a surprising "on-the-line" phenomenon in persona generalization: in-domain and out-of-domain p..."
via Arxivđ¤ Julia Herbst, Simon Roschmann, Quentin Bouniot et al.đ 2026-11-09
⥠Score: 7.0
"The Platonic Representation Hypothesis (PRH) suggests that neural networks are converging toward similar representation spaces. As language models become more capable, their representations are aligning with those of vision models. However, recent work has questioned whether this trend continues to..."
via Arxivđ¤ Zhimin Hu, Chen Jiang, Hanbo Xie et al.đ 2026-11-10
⥠Score: 6.9
"Neural networks are increasingly studied not merely as predictive systems, but as complex systems whose internal computations are themselves objects of scientific inquiry. Recent progress in mechanistic interpretability (MI) has made it possible to identify neurons, features, circuits, population st..."
via Arxivđ¤ Matthew Flathers, Jill Noorily, Phuong Anh Nguyen et al.đ 2026-11-09
⥠Score: 6.9
"Users are asking language models mental health questions and are receiving answers. We audited what 41 language models recommend to users when the same clinical information reaches them from three footings: a clinician describing a patient, a caregiver describing a loved one, or a person describing..."
via Arxivđ¤ Tingcheng Bian, Kangye Gong, Jinchang Luo et al.đ 2026-11-09
⥠Score: 6.9
"Reinforcement learning is now the mainstream paradigm for training LLMs on agentic tasks, yet standard practice keeps only the best-validation checkpoint and discards the rest. We show that this is lossy: in GRPO training of LLM agents, the best-validation checkpoint fails on problems that earlier c..."
via Arxivđ¤ Audrey Huang, Nan Jiang, Sham M. Kakade et al.đ 2026-11-10
⥠Score: 6.8
"Policy optimization is central to language model post-training, yet principled guidance remains limited. We revisit its theoretical foundations to identify algorithms that learn near-optimal policies using total samples and gradient steps governed by the base model's response coverage, a measure of..."
via Arxivđ¤ Jiahang Lin, Wanying Ren, Wanghan Xu et al.đ 2026-11-10
⥠Score: 6.8
"Research agents achieve increasingly strong benchmark scores, yet these scores do not necessarily translate into reliable delivery in ongoing research projects. Many research-agent benchmarks emphasize standalone tasks and relatively self-contained workflows in clean or curated workspaces. However,..."
via Arxivđ¤ Zheng Jiang, Fang Niu, Bo Peng et al.đ 2026-11-09
⥠Score: 6.8
"Large language model agents are increasingly deployed as multi-agent systems rather than as a standalone model operating within a single context. Recent agent harnesses allow the model to instantiate worker agents and to author the executable workflow that coordinates them, shifting the design of sy..."
via Arxivđ¤ Weida Wang, Benteng Chen, Yigeng Jiang et al.đ 2026-11-09
⥠Score: 6.7
"We introduce Sage-Taiyi, a metacognition-inspired architecture that uses accumulated scientific understanding to guide what to investigate next and how to investigate it. Through explicit monitoring and control, the system revises its understanding of evidence, assumptions, and unresolved questions,..."
via Arxivđ¤ Zihan Chen, Lei Nico Zheng, Di Zhuđ 2026-11-10
⥠Score: 6.7
"Users often complain that AI assistants such as ChatGPT are getting worse. Some versions do change, but human evaluation leaves a second possibility untested: the people judging may change. We separate the two on Chatbot Arena, a public platform where people compare answers from two anonymous large..."
via Arxivđ¤ Ziang Liu, James J. Kim, Yijia Dai et al.đ 2026-11-10
⥠Score: 6.7
"LLM agents are increasingly used for open-ended, data-driven scientific discovery, yet most of them evaluate hypotheses pointwise, testing an empirical claim $h$ only against its complement $\neg h$. When evidence rejects $h$, the agent learns that one explanation is unlikely but not which alternati..."
via Arxivđ¤ Muhan Gao, Hangxiao Zhu, Xinqi Su et al.đ 2026-11-10
⥠Score: 6.6
"Research ideation is iterative: researchers refine their ideas step by step based on feedback. Yet existing evaluations obscure this search process, making it difficult to understand how LLMs explore the space of research ideas and what limits their progress. We introduce Re-Search, which adapts the..."
via Arxivđ¤ Yihan Hank Tang, Peng Yang, Jiayi Xin et al.đ 2026-11-09
⥠Score: 6.6
"AI is used in much of the clinical trial pipeline. However, selecting the statistical design of a trial, a crucial step that determines how drug doses are tested on humans and evaluated, has seen almost no AI progress. LLMs appear to be a natural fit given their strengths in biomedical knowledge and..."
via Arxivđ¤ Howard Zhang, Chuck Arvin, Logan Brown et al.đ 2026-11-10
⥠Score: 6.6
"Deploying LLM-based analytics agents in enterprise settings requires evaluation frameworks that can reliably detect failures across complex, multi-tool workflows. We present a three-phase comparative study of three evaluation frameworks, each representing a distinct evaluation paradigm (trace-based..."
via Arxivđ¤ Souhaila Djaffal, Amna Jamal, Gagana Madhugiri Devaraja et al.đ 2026-11-10
⥠Score: 6.4
"Formal world models can support planning, verification, and safety reasoning, but constructing them from open-ended natural-language situations is difficult, particularly in multi-agent settings where several formalizations may be plausible. We study whether five LLMs can generate such world models..."
đŦ "The agent will often just blindly retry the exact same MCP tool call with the exact same arguments"
âĸ "Page content is untrusted input that now flows straight into an agent's context"
"Streaming speech recognizers output a word only after the speaker has finished it, typically 90-400 ms later. For a voice agent every millisecond counts: the sooner the words are outputted, the sooner the agent can start a search or plan its reply. We present Earlybird, a 29M-parameter streaming rec..."
via Arxivđ¤ Zijian Zhang, Rizhen Hu, Jiangweizhi Peng et al.đ 2026-11-10
⥠Score: 6.1
"Reinforcement learning with verifiable rewards (RLVR) produces rich training experience, yet outcome rewards reduce each attempt to a single number and leave most of this information unused. Training-time scaffolds, such as strategies or hints added to the prompt during training, offer a natural way..."