IBM Research published Palette, a method for teaching a small model to build presentation decks in a specified design language.
AWS documented on-behalf-of token exchange between ServiceNow and Bedrock AgentCore Gateway, enabling delegated-auth agent access.
AWS ML Blog published a technical guide on building ultra-low-latency voice AI applications.
OpenAI published a post on rewarding defenders with larger bug bounties, alongside a separate post on preparing its next security sprint.
Ars Technica reports Russia is considering burying data centers underground after Ukrainian drone strikes on infrastructure.
Ars Technica covers Anthropic's new acceptable-use rules barring game cloning via Claude, and argues the restrictions will be ineffective.
The Verge reports that Valve's Steam review pipeline backlog is causing indie developers to miss announced launch dates.
MIT Technology Review published a piece arguing for urgent real-time monitoring of AI agents' actions.
MIT Technology Review published analysis on using reverse game theory and mechanism design to keep AI agents in check.
Hugging Face documented Model Integration Contracts for Transformers, formalizing how new architectures integrate with the library.
Hugging Face published CWE-bench v2, an updated benchmark for common weakness enumeration detection with more coverage, quality, and transparency.
TechCrunch reports data center operators are repurposing old jet engines for on-site power as grid interconnection queues stretch out.
TechCrunch analyses Palantir's quarter and identifies a cautionary signal beneath otherwise strong reported results.
Ted Schlein's Ballistic Ventures is raising $490M across new funds, citing an AI-driven cybersecurity boom.
CoworkOS was posted to Hugging Face Papers as a paper on scaling agent coworkers for real-world automation, with 89 upvotes.
IBM released Granite Switch 4.1.1, framed around updating models like software, alongside the broader IBM Granite 4.1 family release noted on Hacker News.
Spotify published portal-ai-plugins, an open repository of AI plugins. The repo appeared on GitHub trending with modest traction.
AllenAI released StateBench, a benchmark measuring how language models track state changes across long documents. It targets a failure mode relevant to agent memory and long-context reasoning.
DeepSeek's 'deepseek-harness' repository lists 271,204 stars as of 2026-11-04, alongside DeepInfer at 25,242 stars. The metrics come from a scraped lab GitHub listing without direct URL verification.
Andrej Karpathy published nanochat, a minimal cost-bounded ChatGPT-style training/implementation project framed around a $100 budget. The repo gained ~479 stars in a day.
OpenAI published a case study on Cisco using its models to defend networks at machine speed in cybersecurity operations. The post is a vendor case study rather than an independent evaluation.
AWS introduced India-based inference for Moonshot's Kimi K3 model on Amazon Bedrock, expanding regional availability. This enables data-residency-aligned inference for the model.
Nexus Mods, the largest PC game modding site, is implementing AI disclosure requirements for uploaded content. The move establishes platform-level policy on labeling AI-generated or AI-assisted mods.
Google will turn off Search for schools that refuse its AI features, bundling AI adoption with core search access in education. The policy forces institutions to accept AI-driven search or lose search entirely.
DeepMind announced Genie 4, a new generation of its world-simulation model. It advances generative interactive environments used for agent training and simulation.
A supply-chain compromise backdoored the Anthropic VS Code extension distributed through Cursor for nearly 12 hours before remediation. The incident highlights risk in AI coding tool distribution channels.
Inference platform Fal raised at a $16B valuation, double its prior mark, citing surging demand for model serving. The round signals continued investor appetite for inference-layer infrastructure despite bubble concerns.
Anthropic is developing a system to prove which AI model produced a given output, reportedly involving cryptographic proofs of model identity. Reddit discussion references it as 'FlashProve.' It targets the growing problem of verifying model provenance across APIs and deployments.
Google DeepMind announced Gemini 4 Carbon, described as its next frontier model, alongside the previously released Gemini 4 Argon. The release expands Google's Gemini 4 family with a new capability tier.