Policy

Procore replaced its API-driven, once-daily refresh analytics product with OpenSharing on Databricks, doubling data availability, lifting refresh frequency to hourly, and scaling to hundreds of curated tables for construction customers.

Databricks explains that voice agents now sound natural and respond in real time, but real value comes from grounding them in customer context, ontology, and governance to personalize every call and create a learning loop.

OpenAI and ServiceNow are expanding their partnership to integrate OpenAI models and agents into ServiceNow products and connect ServiceNow workflows with ChatGPT and Codex, helping businesses build applications, resolve IT requests and address security issues with AI.

The paper introduces CLAP-AD, a planning paradigm that learns paths via imitation while scoring candidate speeds with a learned function supervised by physically grounded safety constraints. As a plug-in planning head, CLAP-AD enables explicit test-time speed control without speed-conditioned training. It achieves a state-of-the-art Speed-Adherence Score of 82.53 on Bench2Drive-Speed, improves driving score by +28.5 and +32.7 over UniAD and VAD on Bench2Drive, and boosts EPDMS by +6.1 over its backbone on NAVSIM v2.

A study of 10,420 congressional e-newsletters from 2010 to 2025 finds that Democrats and Republicans often attach the same election terms—such as "integrity," "security," and "fraud"—to very different surrounding words. The parties do not need different vocabulary to tell different stories about elections; they only need to place the same words in different company.

Databricks outlines a strategic framework for migrating from Google BigQuery to its open Lakehouse architecture, covering process, technology, and people pillars to enable low-risk, measurable migrations.

OpenAI and Oracle are deepening their partnership to integrate Oracle business applications and data into ChatGPT and power new Oracle offerings with OpenAI models, including AI Security Mesh, SMB Bundle, and App Composer, aiming for an agent-ready enterprise.

The paper introduces Neuromorphic Decision Dynamics (NDD), a decentralized, event-based controller that guarantees safety, liveness, robustness, and scalability for navigation in congested spaces without requiring communication or global information.

This paper presents a modular perception-prediction-planning pipeline that propagates uncertainty as structured, interpretable quantities across modules, decoupling safety guarantees from solver feasibility. In Monte Carlo trials at a four-way unsignalized intersection, the pipeline achieves 0% collisions and 0% solver infeasibility at 12.7 ms per cycle, outperforming deterministic and chance-constrained baselines.

An arXiv paper models academic revision as an admissibility problem under ordered constraints, combining task-relative invariants, contextual constraints and mutable presentation choices with a claim-preservation contract and four governance propositions so that presentation can change without silently altering the research argument.

A new arXiv review surveys non-destructive inspection across four lenses—computational algorithms, physical techniques, industrial applications, and research objectives—charting the shift from manual, analog interpretation to a data-centric NDE 4.0 driven by deep learning and generative models, while flagging scarce standardized open datasets and the AI "black box" problem as key bottlenecks.

The paper proposes deployment failure rates under a declared context distribution as the evidentiary object for use-based AI regulation, studying strategic play among deployers, auditors, and enforcement points. It offers five mechanisms and propositions addressing truthful mixture declarations, unbiased audits, long-lived certification budgets, subgroup guarantees, and misleading incident metrics, tested via five preregistered synthetic studies without evaluating any language model or production system.

This arXiv paper, based on a round-table at COBEM 2025, presents a structured perspective on AI in mechanical engineering, emphasizing that AI should complement rather than replace engineering judgment and proposing principles for responsible adoption including physical consistency, rigorous validation, uncertainty quantification, energy sustainability, human oversight, and ethical responsibility.

Together GPU Clusters auto repair detects GPU, host and fabric faults, picks the lightest fix, and lets you decide per fault and per Xid code which repairs run automatically. Here is how it works.

Databricks blog explains how Lakeflow Connect uses native managed connectors to ingest data from productivity and AI tools like Atlassian, Microsoft, Google, GitHub, OpenAI, and Anthropic into Unity Catalog, eliminating custom pipelines and enabling enterprise knowledge bases, AI agents, and AI spend governance.

CodeRabbit uses GPT-6.1 Sol as the primary model in its security agent to investigate risks across complex codebases, validate findings in isolated sandboxes, and draft fix pull requests for human review.

The media and advertising industry is shifting from counting attention to predicting and proving business outcomes. Databricks argues that a unified, governed data and AI foundation is essential to thrive in the prediction economy, built on four pillars: context, control, cost, and choice.

Fanatics has launched an OpenAI-powered version of Shop the Look that recommends coordinated game-day merchandise. Testing across roughly 30,000 MLB products showed 10% higher engagement than the previous experience.

SEVI introduces a reliability-aware safety intervention framework for real-world robot reinforcement learning, using an evidential safety critic to predict constraint-violation risk and epistemic uncertainty in one forward pass, overriding unsafe or under-supported actions with an offline recovery policy.

StateBridge is a traffic video-language framework trained only on synthetic data that uses phase-aware clip construction, pedestrian and vehicle track summaries, and structured prompts to transfer evidence to real-world traffic, scoring an official S2 of 55.4679 on AI City Challenge 2026 Track 2.

SAGE uses a vision-language model's trajectory as a preference over intact geometric candidates and enforces MUTCD-derived checks, cutting displacement errors and hard-constraint violations on real-world ROADWork data.

An arXiv paper presented at the IJCAI 2026 Workshop on Augmented Democracies studies distributional alignment in large language models: how to make model output distributions match a human reference distribution, and how aligned models generalize to unseen samples, topics, and tasks. The work finds same-domain generalization is achievable, cross-topic generalization depends on topic relationships, and alignment can cause surprising degradations, making generalization inherently unpredictable and unstable.

Researchers introduce $M^3$, an automated red-teaming framework that formalizes multi-turn jailbreaking as a finite-horizon Markov Decision Process. Across two benchmarks and six target models, it improves Attack Success Rates by up to 53% in ASR@5, and even with a lightweight attacker LLM it gains up to 26% in ASR@1 and 34% in ASR@5 over baselines using stronger attackers. It also shows steeper risk escalation over turns and more consistent performance across risk categories without a predefined risk taxonomy.

The paper separates recovery decisions in tool-using AI agents from language-model behavior, treating checkpoint state, permission to dispatch, evidence of external effect, and permission to reconcile as four distinct obligations, and tests them across 336 histories on LangGraph and Temporal under four recovery policies.